baseapi.cpp

Go to the documentation of this file.

/**********************************************************************
 * File:        baseapi.cpp
 * Description: Simple API for calling tesseract.
 * Author:      Ray Smith
 *
 * (C) Copyright 2006, Google Inc.
 ** Licensed under the Apache License, Version 2.0 (the "License");
 ** you may not use this file except in compliance with the License.
 ** You may obtain a copy of the License at
 ** http://www.apache.org/licenses/LICENSE-2.0
 ** Unless required by applicable law or agreed to in writing, software
 ** distributed under the License is distributed on an "AS IS" BASIS,
 ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 ** See the License for the specific language governing permissions and
 ** limitations under the License.
 *
 **********************************************************************/

// Include automatically generated configuration file if running autoconf.
#ifdef HAVE_CONFIG_H
#include "config_auto.h"
#endif

#include "baseapi.h"
#ifdef __linux__
#include <csignal>            // for sigaction, SA_RESETHAND, SIGBUS, SIGFPE
#endif

#if defined(_WIN32)
#if defined(__MINGW32__)
// workaround for stdlib.h with -std=c++11 for _splitpath and _MAX_FNAME
#undef __STRICT_ANSI__
#endif  // __MINGW32__
#include <fcntl.h>
#include <io.h>
#else
#include <dirent.h>            // for closedir, opendir, readdir, DIR, dirent
#include <libgen.h>
#include <sys/types.h>
#include <sys/stat.h>          // for stat, S_IFDIR
#include <unistd.h>
#endif  // _WIN32

#include <clocale>             // for LC_ALL, LC_CTYPE, LC_NUMERIC
#include <cmath>               // for round, M_PI
#include <cstdint>             // for int32_t
#include <cstring>             // for strcmp, strcpy
#include <fstream>             // for size_t
#include <iostream>            // for std::cin
#include <memory>              // for std::unique_ptr
#include <set>                 // for std::pair
#include <vector>              // for std::vector
#include "allheaders.h"        // for pixDestroy, boxCreate, boxaAddBox, box...
#include "blobclass.h"         // for ExtractFontName
#include "boxword.h"           // for BoxWord
#include "config_auto.h"       // for PACKAGE_VERSION
#include "coutln.h"            // for C_OUTLINE_IT, C_OUTLINE_LIST
#include "dawg_cache.h"        // for DawgCache
#include "dict.h"              // for Dict
#include "edgblob.h"           // for extract_edges
#include "elst.h"              // for ELIST_ITERATOR, ELISTIZE, ELISTIZEH
#include "environ.h"           // for l_uint8, FALSE, TRUE
#include "equationdetect.h"    // for EquationDetect
#include "errcode.h"           // for ASSERT_HOST
#include "globaloc.h"          // for SavePixForCrash, signal_exit
#include "helpers.h"           // for IntCastRounded, chomp_string
#include "host.h"              // for BOOL8
#include "imageio.h"           // for IFF_TIFF_G4, IFF_TIFF, IFF_TIFF_G3
#include "intfx.h"             // for INT_FX_RESULT_STRUCT
#include "mutableiterator.h"   // for MutableIterator
#include "normalis.h"          // for kBlnBaselineOffset, kBlnXHeight
#include "ocrclass.h"          // for ETEXT_DESC
#include "openclwrapper.h"     // for PERF_COUNT_END, PERF_COUNT_START, PERF...
#include "osdetect.h"          // for OSResults, OSBestResult, OrientationId...
#include "pageres.h"           // for PAGE_RES_IT, WERD_RES, PAGE_RES, CR_DE...
#include "paragraphs.h"        // for DetectParagraphs
#include "params.h"            // for BoolParam, IntParam, DoubleParam, Stri...
#include "pdblock.h"           // for PDBLK
#include "points.h"            // for FCOORD
#include "polyblk.h"           // for POLY_BLOCK
#include "rect.h"              // for TBOX
#include "renderer.h"          // for TessResultRenderer
#include "resultiterator.h"    // for ResultIterator
#include "stepblob.h"          // for C_BLOB_IT, C_BLOB, C_BLOB_LIST
#include "strngs.h"            // for STRING
#include "tessdatamanager.h"   // for TessdataManager, kTrainedDataSuffix
#include "tesseractclass.h"    // for Tesseract
#include "thresholder.h"       // for ImageThresholder
#include "tprintf.h"           // for tprintf
#include "werd.h"              // for WERD, WERD_IT, W_FUZZY_NON, W_FUZZY_SP

BOOL_VAR(stream_filelist, FALSE, "Stream a filelist from stdin");

namespace tesseract {

const int kMinRectSize = 10;
const char kTesseractReject = '~';
const char kUNLVReject = '~';
const char kUNLVSuspect = '^';
const char* kInputFile = "noname.tif";
const char* kOldVarsFile = "failed_vars.txt";
const int kMaxIntSize = 22;

/* Add all available languages recursively.
*/
static void addAvailableLanguages(const STRING &datadir, const STRING &base,
                                  GenericVector<STRING>* langs)
{
  const STRING base2 = (base.string()[0] == '\0') ? base : base + "/";
  const size_t extlen = sizeof(kTrainedDataSuffix);
#ifdef _WIN32
    WIN32_FIND_DATA data;
    HANDLE handle = FindFirstFile((datadir + base2 + "*").string(), &data);
    if (handle != INVALID_HANDLE_VALUE) {
      BOOL result = TRUE;
      for (; result;) {
        char *name = data.cFileName;
        // Skip '.', '..', and hidden files
        if (name[0] != '.') {
          if ((data.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) ==
              FILE_ATTRIBUTE_DIRECTORY) {
            addAvailableLanguages(datadir, base2 + name, langs);
          } else {
            size_t len = strlen(name);
            if (len > extlen && name[len - extlen] == '.' &&
                strcmp(&name[len - extlen + 1], kTrainedDataSuffix) == 0) {
              name[len - extlen] = '\0';
              langs->push_back(base2 + name);
            }
          }
        }
        result = FindNextFile(handle, &data);
      }
      FindClose(handle);
    }
#else  // _WIN32
  DIR* dir = opendir((datadir + base).string());
  if (dir != nullptr) {
    dirent *de;
    while ((de = readdir(dir))) {
      char *name = de->d_name;
      // Skip '.', '..', and hidden files
      if (name[0] != '.') {
        struct stat st;
        if (stat((datadir + base2 + name).string(), &st) == 0 &&
            (st.st_mode & S_IFDIR) == S_IFDIR) {
          addAvailableLanguages(datadir, base2 + name, langs);
        } else {
          size_t len = strlen(name);
          if (len > extlen && name[len - extlen] == '.' &&
              strcmp(&name[len - extlen + 1], kTrainedDataSuffix) == 0) {
            name[len - extlen] = '\0';
            langs->push_back(base2 + name);
          }
        }
      }
    }
    closedir(dir);
  }
#endif
}

// Compare two STRING values (used for sorting).
static int CompareSTRING(const void* p1, const void* p2) {
  const STRING* s1 = static_cast<const STRING*>(p1);
  const STRING* s2 = static_cast<const STRING*>(p2);
  return strcmp(s1->c_str(), s2->c_str());
}

TessBaseAPI::TessBaseAPI()
    : tesseract_(nullptr),
      osd_tesseract_(nullptr),
      equ_detect_(nullptr),
      reader_(nullptr),
      // Thresholder is initialized to nullptr here, but will be set before use by:
      // A constructor of a derived API,  SetThresholder(), or
      // created implicitly when used in InternalSetImage.
      thresholder_(nullptr),
      paragraph_models_(nullptr),
      block_list_(nullptr),
      page_res_(nullptr),
      input_file_(nullptr),
      output_file_(nullptr),
      datapath_(nullptr),
      language_(nullptr),
      last_oem_requested_(OEM_DEFAULT),
      recognition_done_(false),
      truth_cb_(nullptr),
      rect_left_(0),
      rect_top_(0),
      rect_width_(0),
      rect_height_(0),
      image_width_(0),
      image_height_(0) {
  const char *locale;
  locale = std::setlocale(LC_ALL, nullptr);
  ASSERT_HOST(!strcmp(locale, "C"));
  locale = std::setlocale(LC_CTYPE, nullptr);
  ASSERT_HOST(!strcmp(locale, "C"));
  locale = std::setlocale(LC_NUMERIC, nullptr);
  ASSERT_HOST(!strcmp(locale, "C"));
}

TessBaseAPI::~TessBaseAPI() {
  End();
}

const char* TessBaseAPI::Version() {
  return PACKAGE_VERSION;
}

#ifdef USE_OPENCL
#ifdef USE_DEVICE_SELECTION
#include "opencl_device_selection.h"
#endif
#endif
size_t TessBaseAPI::getOpenCLDevice(void **data) {
#ifdef USE_OPENCL
#ifdef USE_DEVICE_SELECTION
  ds_device device = OpenclDevice::getDeviceSelection();
  if (device.type == DS_DEVICE_OPENCL_DEVICE) {
    *data = new cl_device_id;
    memcpy(*data, &device.oclDeviceID, sizeof(cl_device_id));
    return sizeof(cl_device_id);
  }
#endif
#endif

  *data = nullptr;
  return 0;
}

void TessBaseAPI::CatchSignals() {
#ifdef __linux__
  struct sigaction action;
  memset(&action, 0, sizeof(action));
  action.sa_handler = &signal_exit;
  action.sa_flags = SA_RESETHAND;
  sigaction(SIGSEGV, &action, nullptr);
  sigaction(SIGFPE, &action, nullptr);
  sigaction(SIGBUS, &action, nullptr);
#else
  // Warn API users that an implementation is needed.
  tprintf("CatchSignals has no non-linux implementation!\n");
#endif
}

void TessBaseAPI::SetInputName(const char* name) {
  if (input_file_ == nullptr)
    input_file_ = new STRING(name);
  else
    *input_file_ = name;
}

void TessBaseAPI::SetOutputName(const char* name) {
  if (output_file_ == nullptr)
    output_file_ = new STRING(name);
  else
    *output_file_ = name;
}

bool TessBaseAPI::SetVariable(const char* name, const char* value) {
  if (tesseract_ == nullptr) tesseract_ = new Tesseract;
  return ParamUtils::SetParam(name, value, SET_PARAM_CONSTRAINT_NON_INIT_ONLY,
                              tesseract_->params());
}

bool TessBaseAPI::SetDebugVariable(const char* name, const char* value) {
  if (tesseract_ == nullptr) tesseract_ = new Tesseract;
  return ParamUtils::SetParam(name, value, SET_PARAM_CONSTRAINT_DEBUG_ONLY,
                              tesseract_->params());
}

bool TessBaseAPI::GetIntVariable(const char *name, int *value) const {
  IntParam *p = ParamUtils::FindParam<IntParam>(
      name, GlobalParams()->int_params, tesseract_->params()->int_params);
  if (p == nullptr) return false;
  *value = (int32_t)(*p);
  return true;
}

bool TessBaseAPI::GetBoolVariable(const char *name, bool *value) const {
  BoolParam *p = ParamUtils::FindParam<BoolParam>(
      name, GlobalParams()->bool_params, tesseract_->params()->bool_params);
  if (p == nullptr) return false;
  *value = (BOOL8)(*p);
  return true;
}

const char *TessBaseAPI::GetStringVariable(const char *name) const {
  StringParam *p = ParamUtils::FindParam<StringParam>(
      name, GlobalParams()->string_params, tesseract_->params()->string_params);
  return (p != nullptr) ? p->string() : nullptr;
}

bool TessBaseAPI::GetDoubleVariable(const char *name, double *value) const {
  DoubleParam *p = ParamUtils::FindParam<DoubleParam>(
      name, GlobalParams()->double_params, tesseract_->params()->double_params);
  if (p == nullptr) return false;
  *value = (double)(*p);
  return true;
}

bool TessBaseAPI::GetVariableAsString(const char *name, STRING *val) {
  return ParamUtils::GetParamAsString(name, tesseract_->params(), val);
}

void TessBaseAPI::PrintVariables(FILE *fp) const {
  ParamUtils::PrintParams(fp, tesseract_->params());
}

int TessBaseAPI::Init(const char* datapath, const char* language,
                      OcrEngineMode oem, char **configs, int configs_size,
                      const GenericVector<STRING> *vars_vec,
                      const GenericVector<STRING> *vars_values,
                      bool set_only_non_debug_params) {
  return Init(datapath, 0, language, oem, configs, configs_size, vars_vec,
              vars_values, set_only_non_debug_params, nullptr);
}

// In-memory version reads the traineddata file directly from the given
// data[data_size] array. Also implements the version with a datapath in data,
// flagged by data_size = 0.
int TessBaseAPI::Init(const char* data, int data_size, const char* language,
                      OcrEngineMode oem, char** configs, int configs_size,
                      const GenericVector<STRING>* vars_vec,
                      const GenericVector<STRING>* vars_values,
                      bool set_only_non_debug_params, FileReader reader) {
  PERF_COUNT_START("TessBaseAPI::Init")
  // Default language is "eng".
  if (language == nullptr) language = "eng";
  STRING datapath = data_size == 0 ? data : language;
  // If the datapath, OcrEngineMode or the language have changed - start again.
  // Note that the language_ field stores the last requested language that was
  // initialized successfully, while tesseract_->lang stores the language
  // actually used. They differ only if the requested language was nullptr, in
  // which case tesseract_->lang is set to the Tesseract default ("eng").
  if (tesseract_ != nullptr &&
      (datapath_ == nullptr || language_ == nullptr || *datapath_ != datapath ||
       last_oem_requested_ != oem ||
       (*language_ != language && tesseract_->lang != language))) {
    delete tesseract_;
    tesseract_ = nullptr;
  }
  // PERF_COUNT_SUB("delete tesseract_")
#ifdef USE_OPENCL
  OpenclDevice od;
  od.InitEnv();
#endif
  PERF_COUNT_SUB("OD::InitEnv()")
  bool reset_classifier = true;
  if (tesseract_ == nullptr) {
    reset_classifier = false;
    tesseract_ = new Tesseract;
    if (reader != nullptr) reader_ = reader;
    TessdataManager mgr(reader_);
    if (data_size != 0) {
      mgr.LoadMemBuffer(language, data, data_size);
    }
    if (tesseract_->init_tesseract(
            datapath.string(),
            output_file_ != nullptr ? output_file_->string() : nullptr,
            language, oem, configs, configs_size, vars_vec, vars_values,
            set_only_non_debug_params, &mgr) != 0) {
      return -1;
    }
  }

  PERF_COUNT_SUB("update tesseract_")
  // Update datapath and language requested for the last valid initialization.
  if (datapath_ == nullptr)
    datapath_ = new STRING(datapath);
  else
    *datapath_ = datapath;
  if ((strcmp(datapath_->string(), "") == 0) &&
      (strcmp(tesseract_->datadir.string(), "") != 0))
     *datapath_ = tesseract_->datadir;

  if (language_ == nullptr)
    language_ = new STRING(language);
  else
    *language_ = language;
  last_oem_requested_ = oem;

#ifndef DISABLED_LEGACY_ENGINE
  // PERF_COUNT_SUB("update last_oem_requested_")
  // For same language and datapath, just reset the adaptive classifier.
  if (reset_classifier) {
    tesseract_->ResetAdaptiveClassifier();
    PERF_COUNT_SUB("tesseract_->ResetAdaptiveClassifier()")
  }
#endif  // ndef DISABLED_LEGACY_ENGINE
  PERF_COUNT_END
  return 0;
}

const char* TessBaseAPI::GetInitLanguagesAsString() const {
  return (language_ == nullptr || language_->string() == nullptr) ?
      "" : language_->string();
}

void TessBaseAPI::GetLoadedLanguagesAsVector(
    GenericVector<STRING>* langs) const {
  langs->clear();
  if (tesseract_ != nullptr) {
    langs->push_back(tesseract_->lang);
    int num_subs = tesseract_->num_sub_langs();
    for (int i = 0; i < num_subs; ++i)
      langs->push_back(tesseract_->get_sub_lang(i)->lang);
  }
}

void TessBaseAPI::GetAvailableLanguagesAsVector(
    GenericVector<STRING>* langs) const {
  langs->clear();
  if (tesseract_ != nullptr) {
    addAvailableLanguages(tesseract_->datadir, "", langs);
    langs->sort(CompareSTRING);
  }
}

//TODO(amit): Adapt to lstm
#ifndef DISABLED_LEGACY_ENGINE

int TessBaseAPI::InitLangMod(const char* datapath, const char* language) {
  if (tesseract_ == nullptr)
    tesseract_ = new Tesseract;
  else
    ParamUtils::ResetToDefaults(tesseract_->params());
  TessdataManager mgr;
  return tesseract_->init_tesseract_lm(datapath, nullptr, language, &mgr);
}
#endif  // ndef DISABLED_LEGACY_ENGINE

void TessBaseAPI::InitForAnalysePage() {
  if (tesseract_ == nullptr) {
    tesseract_ = new Tesseract;
    #ifndef DISABLED_LEGACY_ENGINE
    tesseract_->InitAdaptiveClassifier(nullptr);
    #endif
  }
}

void TessBaseAPI::ReadConfigFile(const char* filename) {
  tesseract_->read_config_file(filename, SET_PARAM_CONSTRAINT_NON_INIT_ONLY);
}

void TessBaseAPI::ReadDebugConfigFile(const char* filename) {
  tesseract_->read_config_file(filename, SET_PARAM_CONSTRAINT_DEBUG_ONLY);
}

void TessBaseAPI::SetPageSegMode(PageSegMode mode) {
  if (tesseract_ == nullptr)
    tesseract_ = new Tesseract;
  tesseract_->tessedit_pageseg_mode.set_value(mode);
}

PageSegMode TessBaseAPI::GetPageSegMode() const {
  if (tesseract_ == nullptr)
    return PSM_SINGLE_BLOCK;
  return static_cast<PageSegMode>(
    static_cast<int>(tesseract_->tessedit_pageseg_mode));
}

char* TessBaseAPI::TesseractRect(const unsigned char* imagedata,
                                 int bytes_per_pixel,
                                 int bytes_per_line,
                                 int left, int top,
                                 int width, int height) {
  if (tesseract_ == nullptr || width < kMinRectSize || height < kMinRectSize)
    return nullptr;  // Nothing worth doing.

  // Since this original api didn't give the exact size of the image,
  // we have to invent a reasonable value.
  int bits_per_pixel = bytes_per_pixel == 0 ? 1 : bytes_per_pixel * 8;
  SetImage(imagedata, bytes_per_line * 8 / bits_per_pixel, height + top,
           bytes_per_pixel, bytes_per_line);
  SetRectangle(left, top, width, height);

  return GetUTF8Text();
}

#ifndef DISABLED_LEGACY_ENGINE

void TessBaseAPI::ClearAdaptiveClassifier() {
  if (tesseract_ == nullptr)
    return;
  tesseract_->ResetAdaptiveClassifier();
  tesseract_->ResetDocumentDictionary();
}
#endif  // ndef DISABLED_LEGACY_ENGINE

void TessBaseAPI::SetImage(const unsigned char* imagedata,
                           int width, int height,
                           int bytes_per_pixel, int bytes_per_line) {
  if (InternalSetImage()) {
    thresholder_->SetImage(imagedata, width, height,
                           bytes_per_pixel, bytes_per_line);
    SetInputImage(thresholder_->GetPixRect());
  }
}

void TessBaseAPI::SetSourceResolution(int ppi) {
  if (thresholder_)
    thresholder_->SetSourceYResolution(ppi);
  else
    tprintf("Please call SetImage before SetSourceResolution.\n");
}

void TessBaseAPI::SetImage(Pix* pix) {
  if (InternalSetImage()) {
    if (pixGetSpp(pix) == 4 && pixGetInputFormat(pix) == IFF_PNG) {
      // remove alpha channel from png
      PIX* p1 = pixRemoveAlpha(pix);
      pixSetSpp(p1, 3);
      pix = pixCopy(nullptr, p1);
      pixDestroy(&p1);
    }
    thresholder_->SetImage(pix);
    SetInputImage(thresholder_->GetPixRect());
  }
}

void TessBaseAPI::SetRectangle(int left, int top, int width, int height) {
  if (thresholder_ == nullptr)
    return;
  thresholder_->SetRectangle(left, top, width, height);
  ClearResults();
}

Pix* TessBaseAPI::GetThresholdedImage() {
  if (tesseract_ == nullptr || thresholder_ == nullptr) return nullptr;
  if (tesseract_->pix_binary() == nullptr &&
      !Threshold(tesseract_->mutable_pix_binary())) {
    return nullptr;
  }
  return pixClone(tesseract_->pix_binary());
}

Boxa* TessBaseAPI::GetRegions(Pixa** pixa) {
  return GetComponentImages(RIL_BLOCK, false, pixa, nullptr);
}

Boxa* TessBaseAPI::GetTextlines(const bool raw_image, const int raw_padding,
                                Pixa** pixa, int** blockids, int** paraids) {
  return GetComponentImages(RIL_TEXTLINE, true, raw_image, raw_padding,
                            pixa, blockids, paraids);
}

Boxa* TessBaseAPI::GetStrips(Pixa** pixa, int** blockids) {
  return GetComponentImages(RIL_TEXTLINE, false, pixa, blockids);
}

Boxa* TessBaseAPI::GetWords(Pixa** pixa) {
  return GetComponentImages(RIL_WORD, true, pixa, nullptr);
}

Boxa* TessBaseAPI::GetConnectedComponents(Pixa** pixa) {
  return GetComponentImages(RIL_SYMBOL, true, pixa, nullptr);
}

Boxa* TessBaseAPI::GetComponentImages(PageIteratorLevel level,
                                      bool text_only, bool raw_image,
                                      const int raw_padding,
                                      Pixa** pixa, int** blockids,
                                      int** paraids) {
  PageIterator* page_it = GetIterator();
  if (page_it == nullptr)
    page_it = AnalyseLayout();
  if (page_it == nullptr)
    return nullptr;  // Failed.

  // Count the components to get a size for the arrays.
  int component_count = 0;
  int left, top, right, bottom;

  TessResultCallback<bool>* get_bbox = nullptr;
  if (raw_image) {
    // Get bounding box in original raw image with padding.
    get_bbox = NewPermanentTessCallback(page_it, &PageIterator::BoundingBox,
                                        level, raw_padding,
                                        &left, &top, &right, &bottom);
  } else {
    // Get bounding box from binarized imaged. Note that this could be
    // differently scaled from the original image.
    get_bbox = NewPermanentTessCallback(page_it,
                                        &PageIterator::BoundingBoxInternal,
                                        level, &left, &top, &right, &bottom);
  }
  do {
    if (get_bbox->Run() &&
        (!text_only || PTIsTextType(page_it->BlockType())))
      ++component_count;
  } while (page_it->Next(level));

  Boxa* boxa = boxaCreate(component_count);
  if (pixa != nullptr)
    *pixa = pixaCreate(component_count);
  if (blockids != nullptr)
    *blockids = new int[component_count];
  if (paraids != nullptr)
    *paraids = new int[component_count];

  int blockid = 0;
  int paraid = 0;
  int component_index = 0;
  page_it->Begin();
  do {
    if (get_bbox->Run() &&
        (!text_only || PTIsTextType(page_it->BlockType()))) {
      Box* lbox = boxCreate(left, top, right - left, bottom - top);
      boxaAddBox(boxa, lbox, L_INSERT);
      if (pixa != nullptr) {
        Pix* pix = nullptr;
        if (raw_image) {
          pix = page_it->GetImage(level, raw_padding, GetInputImage(), &left,
                                  &top);
        } else {
          pix = page_it->GetBinaryImage(level);
        }
        pixaAddPix(*pixa, pix, L_INSERT);
        pixaAddBox(*pixa, lbox, L_CLONE);
      }
      if (paraids != nullptr) {
        (*paraids)[component_index] = paraid;
        if (page_it->IsAtFinalElement(RIL_PARA, level))
          ++paraid;
      }
      if (blockids != nullptr) {
        (*blockids)[component_index] = blockid;
        if (page_it->IsAtFinalElement(RIL_BLOCK, level)) {
          ++blockid;
          paraid = 0;
        }
      }
      ++component_index;
    }
  } while (page_it->Next(level));
  delete page_it;
  delete get_bbox;
  return boxa;
}

int TessBaseAPI::GetThresholdedImageScaleFactor() const {
  if (thresholder_ == nullptr) {
    return 0;
  }
  return thresholder_->GetScaleFactor();
}

PageIterator* TessBaseAPI::AnalyseLayout() { return AnalyseLayout(false); }

PageIterator* TessBaseAPI::AnalyseLayout(bool merge_similar_words) {
  if (FindLines() == 0) {
    if (block_list_->empty())
      return nullptr;  // The page was empty.
    page_res_ = new PAGE_RES(merge_similar_words, block_list_, nullptr);
    DetectParagraphs(false);
    return new PageIterator(
        page_res_, tesseract_, thresholder_->GetScaleFactor(),
        thresholder_->GetScaledYResolution(),
        rect_left_, rect_top_, rect_width_, rect_height_);
  }
  return nullptr;
}

int TessBaseAPI::Recognize(ETEXT_DESC* monitor) {
  if (tesseract_ == nullptr)
    return -1;
  if (FindLines() != 0)
    return -1;
  delete page_res_;
  if (block_list_->empty()) {
    page_res_ = new PAGE_RES(false, block_list_,
                             &tesseract_->prev_word_best_choice_);
    return 0; // Empty page.
  }

  tesseract_->SetBlackAndWhitelist();
  recognition_done_ = true;
#ifndef DISABLED_LEGACY_ENGINE
  if (tesseract_->tessedit_resegment_from_line_boxes) {
    page_res_ = tesseract_->ApplyBoxes(*input_file_, true, block_list_);
  } else if (tesseract_->tessedit_resegment_from_boxes) {
    page_res_ = tesseract_->ApplyBoxes(*input_file_, false, block_list_);
  } else
#endif  // ndef DISABLED_LEGACY_ENGINE
  {
    page_res_ = new PAGE_RES(tesseract_->AnyLSTMLang(),
                             block_list_, &tesseract_->prev_word_best_choice_);
  }

  if (page_res_ == nullptr) {
    return -1;
  }

  if (tesseract_->tessedit_train_line_recognizer) {
    tesseract_->TrainLineRecognizer(*input_file_, *output_file_, block_list_);
    tesseract_->CorrectClassifyWords(page_res_);
    return 0;
  }
#ifndef DISABLED_LEGACY_ENGINE
  if (tesseract_->tessedit_make_boxes_from_boxes) {
    tesseract_->CorrectClassifyWords(page_res_);
    return 0;
  }
#endif  // ndef DISABLED_LEGACY_ENGINE

  if (truth_cb_ != nullptr) {
    tesseract_->wordrec_run_blamer.set_value(true);
    PageIterator *page_it = new PageIterator(
            page_res_, tesseract_, thresholder_->GetScaleFactor(),
            thresholder_->GetScaledYResolution(),
            rect_left_, rect_top_, rect_width_, rect_height_);
    truth_cb_->Run(tesseract_->getDict().getUnicharset(),
                   image_height_, page_it, this->tesseract()->pix_grey());
    delete page_it;
  }

  int result = 0;
  if (tesseract_->interactive_display_mode) {
    #ifndef GRAPHICS_DISABLED
    tesseract_->pgeditor_main(rect_width_, rect_height_, page_res_);
    #endif  // GRAPHICS_DISABLED
    // The page_res is invalid after an interactive session, so cleanup
    // in a way that lets us continue to the next page without crashing.
    delete page_res_;
    page_res_ = nullptr;
    return -1;
  #ifndef DISABLED_LEGACY_ENGINE
  } else if (tesseract_->tessedit_train_from_boxes) {
    STRING fontname;
    ExtractFontName(*output_file_, &fontname);
    tesseract_->ApplyBoxTraining(fontname, page_res_);
  } else if (tesseract_->tessedit_ambigs_training) {
    FILE *training_output_file = tesseract_->init_recog_training(*input_file_);
    // OCR the page segmented into words by tesseract.
    tesseract_->recog_training_segmented(
        *input_file_, page_res_, monitor, training_output_file);
    fclose(training_output_file);
  #endif  // ndef DISABLED_LEGACY_ENGINE
  } else {
    // Now run the main recognition.
    bool wait_for_text = true;
    GetBoolVariable("paragraph_text_based", &wait_for_text);
    if (!wait_for_text) DetectParagraphs(false);
    if (tesseract_->recog_all_words(page_res_, monitor, nullptr, nullptr, 0)) {
      if (wait_for_text) DetectParagraphs(true);
    } else {
      result = -1;
    }
  }
  return result;
}

#ifndef DISABLED_LEGACY_ENGINE

int TessBaseAPI::RecognizeForChopTest(ETEXT_DESC* monitor) {
  if (tesseract_ == nullptr)
    return -1;
  if (thresholder_ == nullptr || thresholder_->IsEmpty()) {
    tprintf("Please call SetImage before attempting recognition.\n");
    return -1;
  }
  if (page_res_ != nullptr)
    ClearResults();
  if (FindLines() != 0)
    return -1;
  // Additional conditions under which chopper test cannot be run
  if (tesseract_->interactive_display_mode) return -1;

  recognition_done_ = true;

  page_res_ = new PAGE_RES(false, block_list_,
                           &(tesseract_->prev_word_best_choice_));

  PAGE_RES_IT page_res_it(page_res_);

  while (page_res_it.word() != nullptr) {
    WERD_RES *word_res = page_res_it.word();
    GenericVector<TBOX> boxes;
    tesseract_->MaximallyChopWord(boxes, page_res_it.block()->block,
                                  page_res_it.row()->row, word_res);
    page_res_it.forward();
  }
  return 0;
}
#endif  // ndef DISABLED_LEGACY_ENGINE

// Takes ownership of the input pix.
void TessBaseAPI::SetInputImage(Pix* pix) { tesseract_->set_pix_original(pix); }

Pix* TessBaseAPI::GetInputImage() { return tesseract_->pix_original(); }

const char * TessBaseAPI::GetInputName() {
  if (input_file_)
    return input_file_->c_str();
  return nullptr;
}

const char *  TessBaseAPI::GetDatapath() {
  return tesseract_->datadir.c_str();
}

int TessBaseAPI::GetSourceYResolution() {
  return thresholder_->GetSourceYResolution();
}

// If flist exists, get data from there. Otherwise get data from buf.
// Seems convoluted, but is the easiest way I know of to meet multiple
// goals. Support streaming from stdin, and also work on platforms
// lacking fmemopen.
bool TessBaseAPI::ProcessPagesFileList(FILE *flist,
                                       STRING *buf,
                                       const char* retry_config,
                                       int timeout_millisec,
                                       TessResultRenderer* renderer,
                                       int tessedit_page_number) {
  if (!flist && !buf) return false;
  int page = (tessedit_page_number >= 0) ? tessedit_page_number : 0;
  char pagename[MAX_PATH];

  GenericVector<STRING> lines;
  if (!flist) {
    buf->split('\n', &lines);
    if (lines.empty()) return false;
  }

  // Skip to the requested page number.
  for (int i = 0; i < page; i++) {
    if (flist) {
      if (fgets(pagename, sizeof(pagename), flist) == nullptr) break;
    }
  }

  // Begin producing output
  if (renderer && !renderer->BeginDocument(unknown_title_)) {
    return false;
  }

  // Loop over all pages - or just the requested one
  while (true) {
    if (flist) {
      if (fgets(pagename, sizeof(pagename), flist) == nullptr) break;
    } else {
      if (page >= lines.size()) break;
      snprintf(pagename, sizeof(pagename), "%s", lines[page].c_str());
    }
    chomp_string(pagename);
    Pix *pix = pixRead(pagename);
    if (pix == nullptr) {
      tprintf("Image file %s cannot be read!\n", pagename);
      return false;
    }
    tprintf("Page %d : %s\n", page, pagename);
    bool r = ProcessPage(pix, page, pagename, retry_config,
                         timeout_millisec, renderer);
    pixDestroy(&pix);
    if (!r) return false;
    if (tessedit_page_number >= 0) break;
    ++page;
  }

  // Finish producing output
  if (renderer && !renderer->EndDocument()) {
    return false;
  }
  return true;
}

bool TessBaseAPI::ProcessPagesMultipageTiff(const l_uint8 *data,
                                            size_t size,
                                            const char* filename,
                                            const char* retry_config,
                                            int timeout_millisec,
                                            TessResultRenderer* renderer,
                                            int tessedit_page_number) {
#ifndef ANDROID_BUILD
  Pix *pix = nullptr;
  int page = (tessedit_page_number >= 0) ? tessedit_page_number : 0;
  size_t offset = 0;
  for (; ; ++page) {
    if (tessedit_page_number >= 0)
      page = tessedit_page_number;
    pix = (data) ? pixReadMemFromMultipageTiff(data, size, &offset)
                 : pixReadFromMultipageTiff(filename, &offset);
    if (pix == nullptr) break;
    tprintf("Page %d\n", page + 1);
    char page_str[kMaxIntSize];
    snprintf(page_str, kMaxIntSize - 1, "%d", page);
    SetVariable("applybox_page", page_str);
    bool r = ProcessPage(pix, page, filename, retry_config,
                           timeout_millisec, renderer);
    pixDestroy(&pix);
    if (!r) return false;
    if (tessedit_page_number >= 0) break;
    if (!offset) break;
  }
  return true;
#else
  return false;
#endif
}

// Master ProcessPages calls ProcessPagesInternal and then does any post-
// processing required due to being in a training mode.
bool TessBaseAPI::ProcessPages(const char* filename, const char* retry_config,
                               int timeout_millisec,
                               TessResultRenderer* renderer) {
  bool result =
      ProcessPagesInternal(filename, retry_config, timeout_millisec, renderer);
  #ifndef DISABLED_LEGACY_ENGINE
  if (result) {
    if (tesseract_->tessedit_train_from_boxes &&
        !tesseract_->WriteTRFile(*output_file_)) {
      tprintf("Write of TR file failed: %s\n", output_file_->string());
      return false;
    }
  }
  #endif  // ndef DISABLED_LEGACY_ENGINE
  return result;
}

// In the ideal scenario, Tesseract will start working on data as soon
// as it can. For example, if you stream a filelist through stdin, we
// should start the OCR process as soon as the first filename is
// available. This is particularly useful when hooking Tesseract up to
// slow hardware such as a book scanning machine.
//
// Unfortunately there are tradeoffs. You can't seek on stdin. That
// makes automatic detection of datatype (TIFF? filelist? PNG?)
// impractical.  So we support a command line flag to explicitly
// identify the scenario that really matters: filelists on
// stdin. We'll still do our best if the user likes pipes.
bool TessBaseAPI::ProcessPagesInternal(const char* filename,
                                       const char* retry_config,
                                       int timeout_millisec,
                                       TessResultRenderer* renderer) {
  PERF_COUNT_START("ProcessPages")
  bool stdInput = !strcmp(filename, "stdin") || !strcmp(filename, "-");
  if (stdInput) {
#ifdef WIN32
    if (_setmode(_fileno(stdin), _O_BINARY) == -1)
      tprintf("ERROR: cin to binary: %s", strerror(errno));
#endif  // WIN32
  }

  if (stream_filelist) {
    return ProcessPagesFileList(stdin, nullptr, retry_config,
                                timeout_millisec, renderer,
                                tesseract_->tessedit_page_number);
  }

  // At this point we are officially in autodection territory.
  // That means any data in stdin must be buffered, to make it
  // seekable.
  std::string buf;
  const l_uint8 *data = nullptr;
  if (stdInput) {
    buf.assign((std::istreambuf_iterator<char>(std::cin)),
               (std::istreambuf_iterator<char>()));
    data = reinterpret_cast<const l_uint8 *>(buf.data());
  } else {
    // Check whether the input file can be read.
    if (FILE* file = fopen(filename, "rb")) {
      fclose(file);
    } else {
      fprintf(stderr, "Error, cannot read input file %s: %s\n",
              filename, strerror(errno));
      return false;
    }
  }

  // Here is our autodetection
  int format;
  int r = (stdInput) ?
      findFileFormatBuffer(data, &format) :
      findFileFormat(filename, &format);

  // Maybe we have a filelist
  if (r != 0 || format == IFF_UNKNOWN) {
    STRING s;
    if (stdInput) {
      s = buf.c_str();
    } else {
      std::ifstream t(filename);
      std::string u((std::istreambuf_iterator<char>(t)),
                    std::istreambuf_iterator<char>());
      s = u.c_str();
    }
    return ProcessPagesFileList(nullptr, &s, retry_config,
                                timeout_millisec, renderer,
                                tesseract_->tessedit_page_number);
  }

  // Maybe we have a TIFF which is potentially multipage
  bool tiff = (format == IFF_TIFF || format == IFF_TIFF_PACKBITS ||
               format == IFF_TIFF_RLE || format == IFF_TIFF_G3 ||
               format == IFF_TIFF_G4 || format == IFF_TIFF_LZW ||
               format == IFF_TIFF_ZIP);

  // Fail early if we can, before producing any output
  Pix *pix = nullptr;
  if (!tiff) {
    pix = (stdInput) ? pixReadMem(data, buf.size()) : pixRead(filename);
    if (pix == nullptr) {
      return false;
    }
  }

  // Begin the output
  if (renderer && !renderer->BeginDocument(unknown_title_)) {
    pixDestroy(&pix);
    return false;
  }

  // Produce output
  r = (tiff) ?
      ProcessPagesMultipageTiff(data, buf.size(), filename, retry_config,
                                timeout_millisec, renderer,
                                tesseract_->tessedit_page_number) :
      ProcessPage(pix, 0, filename, retry_config,
                  timeout_millisec, renderer);

  // Clean up memory as needed
  pixDestroy(&pix);

  // End the output
  if (!r || (renderer && !renderer->EndDocument())) {
    return false;
  }
  PERF_COUNT_END
  return true;
}

bool TessBaseAPI::ProcessPage(Pix* pix, int page_index, const char* filename,
                              const char* retry_config, int timeout_millisec,
                              TessResultRenderer* renderer) {
  PERF_COUNT_START("ProcessPage")
  SetInputName(filename);
  SetImage(pix);
  bool failed = false;

  if (tesseract_->tessedit_pageseg_mode == PSM_AUTO_ONLY) {
    // Disabled character recognition
    PageIterator* it = AnalyseLayout();

    if (it == nullptr) {
      failed = true;
    } else {
      delete it;
    }
  } else if (tesseract_->tessedit_pageseg_mode == PSM_OSD_ONLY) {
    failed = FindLines() != 0;
  } else if (timeout_millisec > 0) {
    // Running with a timeout.
    ETEXT_DESC monitor;
    monitor.cancel = nullptr;
    monitor.cancel_this = nullptr;
    monitor.set_deadline_msecs(timeout_millisec);

    // Now run the main recognition.
    failed = Recognize(&monitor) < 0;
  } else {
    // Normal layout and character recognition with no timeout.
    failed = Recognize(nullptr) < 0;
  }

  if (tesseract_->tessedit_write_images) {
#ifndef ANDROID_BUILD
    Pix* page_pix = GetThresholdedImage();
    pixWrite("tessinput.tif", page_pix, IFF_TIFF_G4);
#endif  // ANDROID_BUILD
  }

  if (failed && retry_config != nullptr && retry_config[0] != '\0') {
    // Save current config variables before switching modes.
    FILE* fp = fopen(kOldVarsFile, "wb");
    if (fp == nullptr) {
      tprintf("Error, failed to open file \"%s\"\n", kOldVarsFile);
    } else {
      PrintVariables(fp);
      fclose(fp);
    }
    // Switch to alternate mode for retry.
    ReadConfigFile(retry_config);
    SetImage(pix);
    Recognize(nullptr);
    // Restore saved config variables.
    ReadConfigFile(kOldVarsFile);
  }

  if (renderer && !failed) {
    failed = !renderer->AddImage(this);
  }

  PERF_COUNT_END
  return !failed;
}

LTRResultIterator* TessBaseAPI::GetLTRIterator() {
  if (tesseract_ == nullptr || page_res_ == nullptr)
    return nullptr;
  return new LTRResultIterator(
      page_res_, tesseract_,
      thresholder_->GetScaleFactor(), thresholder_->GetScaledYResolution(),
      rect_left_, rect_top_, rect_width_, rect_height_);
}

ResultIterator* TessBaseAPI::GetIterator() {
  if (tesseract_ == nullptr || page_res_ == nullptr)
    return nullptr;
  return ResultIterator::StartOfParagraph(LTRResultIterator(
      page_res_, tesseract_,
      thresholder_->GetScaleFactor(), thresholder_->GetScaledYResolution(),
      rect_left_, rect_top_, rect_width_, rect_height_));
}

MutableIterator* TessBaseAPI::GetMutableIterator() {
  if (tesseract_ == nullptr || page_res_ == nullptr)
    return nullptr;
  return new MutableIterator(page_res_, tesseract_,
                             thresholder_->GetScaleFactor(),
                             thresholder_->GetScaledYResolution(),
                             rect_left_, rect_top_, rect_width_, rect_height_);
}

char* TessBaseAPI::GetUTF8Text() {
  if (tesseract_ == nullptr ||
      (!recognition_done_ && Recognize(nullptr) < 0))
    return nullptr;
  STRING text("");
  ResultIterator *it = GetIterator();
  do {
    if (it->Empty(RIL_PARA)) continue;
    const std::unique_ptr<const char[]> para_text(it->GetUTF8Text(RIL_PARA));
    text += para_text.get();
  } while (it->Next(RIL_PARA));
  char* result = new char[text.length() + 1];
  strncpy(result, text.string(), text.length() + 1);
  delete it;
  return result;
}

static tesseract::Orientation GetBlockTextOrientation(const PageIterator *it) {
  tesseract::Orientation orientation;
  tesseract::WritingDirection writing_direction;
  tesseract::TextlineOrder textline_order;
  float deskew_angle;
  it->Orientation(&orientation, &writing_direction, &textline_order,
                  &deskew_angle);
  return orientation;
}

static void AddBaselineCoordsTohOCR(const PageIterator *it,
                                    PageIteratorLevel level,
                                    STRING* hocr_str) {
  tesseract::Orientation orientation = GetBlockTextOrientation(it);
  if (orientation != ORIENTATION_PAGE_UP) {
    hocr_str->add_str_int("; textangle ", 360 - orientation * 90);
    return;
  }

  int left, top, right, bottom;
  it->BoundingBox(level, &left, &top, &right, &bottom);

  // Try to get the baseline coordinates at this level.
  int x1, y1, x2, y2;
  if (!it->Baseline(level, &x1, &y1, &x2, &y2))
    return;
  // Following the description of this field of the hOCR spec, we convert the
  // baseline coordinates so that "the bottom left of the bounding box is the
  // origin".
  x1 -= left;
  x2 -= left;
  y1 -= bottom;
  y2 -= bottom;

  // Now fit a line through the points so we can extract coefficients for the
  // equation:  y = p1 x + p0
  double p1 = 0;
  double p0 = 0;
  if (x1 == x2) {
    // Problem computing the polynomial coefficients.
    return;
  }
  p1 = (y2 - y1) / static_cast<double>(x2 - x1);
  p0 = y1 - static_cast<double>(p1 * x1);

  hocr_str->add_str_double("; baseline ", round(p1 * 1000.0) / 1000.0);
  hocr_str->add_str_double(" ", round(p0 * 1000.0) / 1000.0);
}

static void AddIdTohOCR(STRING* hocr_str, const std::string base, int num1,
                        int num2) {
  const size_t BUFSIZE = 64;
  char id_buffer[BUFSIZE];
  if (num2 >= 0) {
    snprintf(id_buffer, BUFSIZE - 1, "%s_%d_%d", base.c_str(), num1, num2);
  } else {
    snprintf(id_buffer, BUFSIZE - 1, "%s_%d", base.c_str(), num1);
  }
  id_buffer[BUFSIZE - 1] = '\0';
  *hocr_str += " id='";
  *hocr_str += id_buffer;
  *hocr_str += "'";
}

static void AddIdTohOCR(STRING* hocr_str, const std::string base, int num1,
  int num2, int num3) {
  const size_t BUFSIZE = 64;
  char id_buffer[BUFSIZE];
  snprintf(id_buffer, BUFSIZE - 1, "%s_%d_%d_%d", base.c_str(), num1, num2,num3);
  id_buffer[BUFSIZE - 1] = '\0';
  *hocr_str += " id='";
  *hocr_str += id_buffer;
  *hocr_str += "'";
}

static void AddBoxTohOCR(const ResultIterator* it, PageIteratorLevel level,
                         STRING* hocr_str) {
  int left, top, right, bottom;
  it->BoundingBox(level, &left, &top, &right, &bottom);
  // This is the only place we use double quotes instead of single quotes,
  // but it may too late to change for consistency
  hocr_str->add_str_int(" title=\"bbox ", left);
  hocr_str->add_str_int(" ", top);
  hocr_str->add_str_int(" ", right);
  hocr_str->add_str_int(" ", bottom);
  // Add baseline coordinates & heights for textlines only.
  if (level == RIL_TEXTLINE) {
    AddBaselineCoordsTohOCR(it, level, hocr_str);
    // add custom height measures
    float row_height, descenders, ascenders;  // row attributes
    it->RowAttributes(&row_height, &descenders, &ascenders);
    // TODO(rays): Do we want to limit these to a single decimal place?
    hocr_str->add_str_double("; x_size ", row_height);
    hocr_str->add_str_double("; x_descenders ", descenders * -1);
    hocr_str->add_str_double("; x_ascenders ", ascenders);
  }
  *hocr_str += "\">";
}

static void AddBoxToTSV(const PageIterator* it, PageIteratorLevel level,
                        STRING* hocr_str) {
  int left, top, right, bottom;
  it->BoundingBox(level, &left, &top, &right, &bottom);
  hocr_str->add_str_int("\t", left);
  hocr_str->add_str_int("\t", top);
  hocr_str->add_str_int("\t", right - left);
  hocr_str->add_str_int("\t", bottom - top);
}

char* TessBaseAPI::GetHOCRText(int page_number) {
  return GetHOCRText(nullptr, page_number);
}

char* TessBaseAPI::GetHOCRText(ETEXT_DESC* monitor, int page_number) {
  if (tesseract_ == nullptr || (page_res_ == nullptr && Recognize(monitor) < 0))
    return nullptr;

  int lcnt = 1, bcnt = 1, pcnt = 1, wcnt = 1, tcnt = 1, gcnt = 1;
  int page_id = page_number + 1;  // hOCR uses 1-based page numbers.
  bool para_is_ltr = true;        // Default direction is LTR
  const char* paragraph_lang = nullptr;
  bool font_info = false;
  GetBoolVariable("hocr_font_info", &font_info);

  STRING hocr_str("");

  if (input_file_ == nullptr)
      SetInputName(nullptr);

#ifdef _WIN32
  // convert input name from ANSI encoding to utf-8
  int str16_len =
      MultiByteToWideChar(CP_ACP, 0, input_file_->string(), -1, nullptr, 0);
  wchar_t *uni16_str = new WCHAR[str16_len];
  str16_len = MultiByteToWideChar(CP_ACP, 0, input_file_->string(), -1,
                                  uni16_str, str16_len);
  int utf8_len = WideCharToMultiByte(CP_UTF8, 0, uni16_str, str16_len, nullptr, 0,
                                     nullptr, nullptr);
  char *utf8_str = new char[utf8_len];
  WideCharToMultiByte(CP_UTF8, 0, uni16_str, str16_len, utf8_str,
                      utf8_len, nullptr, nullptr);
  *input_file_ = utf8_str;
  delete[] uni16_str;
  delete[] utf8_str;
#endif

  hocr_str += "  <div class='ocr_page'";
  AddIdTohOCR(&hocr_str, "page", page_id, -1);
  hocr_str += " title='image \"";
  if (input_file_) {
    hocr_str += HOcrEscape(input_file_->string());
  } else {
    hocr_str += "unknown";
  }
  hocr_str.add_str_int("\"; bbox ", rect_left_);
  hocr_str.add_str_int(" ", rect_top_);
  hocr_str.add_str_int(" ", rect_width_);
  hocr_str.add_str_int(" ", rect_height_);
  hocr_str.add_str_int("; ppageno ", page_number);
  hocr_str += "'>\n";

  ResultIterator *res_it = GetIterator();
  while (!res_it->Empty(RIL_BLOCK)) {
    if (res_it->Empty(RIL_WORD)) {
      res_it->Next(RIL_WORD);
      continue;
    }

    // Open any new block/paragraph/textline.
    if (res_it->IsAtBeginningOf(RIL_BLOCK)) {
      para_is_ltr = true;  // reset to default direction
      hocr_str += "   <div class='ocr_carea'";
      AddIdTohOCR(&hocr_str, "block", page_id, bcnt);
      AddBoxTohOCR(res_it, RIL_BLOCK, &hocr_str);
    }
    if (res_it->IsAtBeginningOf(RIL_PARA)) {
      hocr_str += "\n    <p class='ocr_par'";
      para_is_ltr = res_it->ParagraphIsLtr();
      if (!para_is_ltr) {
        hocr_str += " dir='rtl'";
      }
      AddIdTohOCR(&hocr_str, "par", page_id, pcnt);
      paragraph_lang = res_it->WordRecognitionLanguage();
      if (paragraph_lang) {
        hocr_str += " lang='";
        hocr_str += paragraph_lang;
        hocr_str += "'";
      }
      AddBoxTohOCR(res_it, RIL_PARA, &hocr_str);
    }
    if (res_it->IsAtBeginningOf(RIL_TEXTLINE)) {
      hocr_str += "\n     <span class='ocr_line'";
      AddIdTohOCR(&hocr_str, "line", page_id, lcnt);
      AddBoxTohOCR(res_it, RIL_TEXTLINE, &hocr_str);
    }

    // Now, process the word...
    std::vector<std::vector<std::pair<const char*, float>>>* confidencemap = nullptr;
    if (tesseract_->lstm_choice_mode) {
      confidencemap = res_it->GetBestLSTMSymbolChoices();
    }
    hocr_str += "\n      <span class='ocrx_word'";
    AddIdTohOCR(&hocr_str, "word", page_id, wcnt);
    int left, top, right, bottom;
    bool bold, italic, underlined, monospace, serif, smallcaps;
    int pointsize, font_id;
    const char *font_name;
    res_it->BoundingBox(RIL_WORD, &left, &top, &right, &bottom);
    font_name = res_it->WordFontAttributes(&bold, &italic, &underlined,
                                           &monospace, &serif, &smallcaps,
                                           &pointsize, &font_id);
    hocr_str.add_str_int(" title='bbox ", left);
    hocr_str.add_str_int(" ", top);
    hocr_str.add_str_int(" ", right);
    hocr_str.add_str_int(" ", bottom);
    hocr_str.add_str_int("; x_wconf ", res_it->Confidence(RIL_WORD));
    if (font_info) {
      if (font_name) {
        hocr_str += "; x_font ";
        hocr_str += HOcrEscape(font_name);
      }
      hocr_str.add_str_int("; x_fsize ", pointsize);
    }
    hocr_str += "'";
    const char* lang = res_it->WordRecognitionLanguage();
    if (lang && (!paragraph_lang || strcmp(lang, paragraph_lang))) {
      hocr_str += " lang='";
      hocr_str += lang;
      hocr_str += "'";
    }
    switch (res_it->WordDirection()) {
      // Only emit direction if different from current paragraph direction
      case DIR_LEFT_TO_RIGHT:
        if (!para_is_ltr) hocr_str += " dir='ltr'";
        break;
      case DIR_RIGHT_TO_LEFT:
        if (para_is_ltr) hocr_str += " dir='rtl'";
        break;
      case DIR_MIX:
      case DIR_NEUTRAL:
      default:  // Do nothing.
        break;
    }
    hocr_str += ">";
    bool last_word_in_line = res_it->IsAtFinalElement(RIL_TEXTLINE, RIL_WORD);
    bool last_word_in_para = res_it->IsAtFinalElement(RIL_PARA, RIL_WORD);
    bool last_word_in_block = res_it->IsAtFinalElement(RIL_BLOCK, RIL_WORD);
    if (bold) hocr_str += "<strong>";
    if (italic) hocr_str += "<em>";
    do {
      const std::unique_ptr<const char[]> grapheme(
          res_it->GetUTF8Text(RIL_SYMBOL));
      if (grapheme && grapheme[0] != 0) {
        hocr_str += HOcrEscape(grapheme.get());
      }
      res_it->Next(RIL_SYMBOL);
    } while (!res_it->Empty(RIL_BLOCK) && !res_it->IsAtBeginningOf(RIL_WORD));
    if (italic) hocr_str += "</em>";
    if (bold) hocr_str += "</strong>";
    // If the lstm choice mode is required it is added here
    if (tesseract_->lstm_choice_mode == 1 && confidencemap != nullptr) {
      for (size_t i = 0; i < confidencemap->size(); i++) {
        hocr_str += "\n       <span class='ocrx_cinfo'";
        AddIdTohOCR(&hocr_str, "timestep", page_id, wcnt, tcnt);
        hocr_str += ">";
        std::vector<std::pair<const char*, float>> timestep = (*confidencemap)[i];
        for (std::pair<const char*, float> conf : timestep) {
          hocr_str += "<span class='ocr_glyph'";
          AddIdTohOCR(&hocr_str, "choice", page_id, wcnt, gcnt);
          hocr_str.add_str_int(" title='x_confs ", int(conf.second * 100));
          hocr_str += "'";
          hocr_str += ">";
          hocr_str += conf.first;
          hocr_str += "</span>";
          gcnt++;
        }
        hocr_str += "</span>";
        tcnt++;
      }
    } else if (tesseract_->lstm_choice_mode == 2 && confidencemap != nullptr) {
      for (size_t i = 0; i < confidencemap->size(); i++) {
        std::vector<std::pair<const char*, float>> timestep = (*confidencemap)[i];
        if (timestep.size() > 0) {
          hocr_str += "\n       <span class='ocrx_cinfo'";
          AddIdTohOCR(&hocr_str, "lstm_choices", page_id, wcnt, tcnt);
          hocr_str += " chosen='";
          hocr_str += timestep[0].first;
          hocr_str += "'>";
          for (size_t j = 1; j < timestep.size(); j++) {
            hocr_str += "<span class='ocr_glyph'";
            AddIdTohOCR(&hocr_str, "choice", page_id, wcnt, gcnt);
            hocr_str.add_str_int(" title='x_confs ", int(timestep[j].second * 100));
            hocr_str += "'";
            hocr_str += ">";
            hocr_str += timestep[j].first;
            hocr_str += "</span>";
            gcnt++;
          }
          hocr_str += "</span>";
          tcnt++;
        }
      }
    }
    hocr_str += "</span>";
    tcnt = 1;
    gcnt = 1;
    wcnt++;
    // Close any ending block/paragraph/textline.
    if (last_word_in_line) {
      hocr_str += "\n     </span>";
      lcnt++;
    }
    if (last_word_in_para) {
      hocr_str += "\n    </p>\n";
      pcnt++;
      para_is_ltr = true;  // back to default direction
    }
    if (last_word_in_block) {
      hocr_str += "   </div>\n";
      bcnt++;
    }
  }
  hocr_str += "  </div>\n";

  char *ret = new char[hocr_str.length() + 1];
  strcpy(ret, hocr_str.string());
  delete res_it;
  return ret;
}

char* TessBaseAPI::GetTSVText(int page_number) {
  if (tesseract_ == nullptr || (page_res_ == nullptr && Recognize(nullptr) < 0))
    return nullptr;

  int lcnt = 1, bcnt = 1, pcnt = 1, wcnt = 1;
  int page_id = page_number + 1;  // we use 1-based page numbers.

  STRING tsv_str("");

  int page_num = page_id;
  int block_num = 0;
  int par_num = 0;
  int line_num = 0;
  int word_num = 0;

  tsv_str.add_str_int("1\t", page_num);  // level 1 - page
  tsv_str.add_str_int("\t", block_num);
  tsv_str.add_str_int("\t", par_num);
  tsv_str.add_str_int("\t", line_num);
  tsv_str.add_str_int("\t", word_num);
  tsv_str.add_str_int("\t", rect_left_);
  tsv_str.add_str_int("\t", rect_top_);
  tsv_str.add_str_int("\t", rect_width_);
  tsv_str.add_str_int("\t", rect_height_);
  tsv_str += "\t-1\t\n";

  ResultIterator* res_it = GetIterator();
  while (!res_it->Empty(RIL_BLOCK)) {
    if (res_it->Empty(RIL_WORD)) {
      res_it->Next(RIL_WORD);
      continue;
    }

    // Add rows for any new block/paragraph/textline.
    if (res_it->IsAtBeginningOf(RIL_BLOCK)) {
      block_num++;
      par_num = 0;
      line_num = 0;
      word_num = 0;
      tsv_str.add_str_int("2\t", page_num);  // level 2 - block
      tsv_str.add_str_int("\t", block_num);
      tsv_str.add_str_int("\t", par_num);
      tsv_str.add_str_int("\t", line_num);
      tsv_str.add_str_int("\t", word_num);
      AddBoxToTSV(res_it, RIL_BLOCK, &tsv_str);
      tsv_str += "\t-1\t\n";  // end of row for block
    }
    if (res_it->IsAtBeginningOf(RIL_PARA)) {
      par_num++;
      line_num = 0;
      word_num = 0;
      tsv_str.add_str_int("3\t", page_num);  // level 3 - paragraph
      tsv_str.add_str_int("\t", block_num);
      tsv_str.add_str_int("\t", par_num);
      tsv_str.add_str_int("\t", line_num);
      tsv_str.add_str_int("\t", word_num);
      AddBoxToTSV(res_it, RIL_PARA, &tsv_str);
      tsv_str += "\t-1\t\n";  // end of row for para
    }
    if (res_it->IsAtBeginningOf(RIL_TEXTLINE)) {
      line_num++;
      word_num = 0;
      tsv_str.add_str_int("4\t", page_num);  // level 4 - line
      tsv_str.add_str_int("\t", block_num);
      tsv_str.add_str_int("\t", par_num);
      tsv_str.add_str_int("\t", line_num);
      tsv_str.add_str_int("\t", word_num);
      AddBoxToTSV(res_it, RIL_TEXTLINE, &tsv_str);
      tsv_str += "\t-1\t\n";  // end of row for line
    }

    // Now, process the word...
    int left, top, right, bottom;
    res_it->BoundingBox(RIL_WORD, &left, &top, &right, &bottom);
    word_num++;
    tsv_str.add_str_int("5\t", page_num);  // level 5 - word
    tsv_str.add_str_int("\t", block_num);
    tsv_str.add_str_int("\t", par_num);
    tsv_str.add_str_int("\t", line_num);
    tsv_str.add_str_int("\t", word_num);
    tsv_str.add_str_int("\t", left);
    tsv_str.add_str_int("\t", top);
    tsv_str.add_str_int("\t", right - left);
    tsv_str.add_str_int("\t", bottom - top);
    tsv_str.add_str_int("\t", res_it->Confidence(RIL_WORD));
    tsv_str += "\t";

    // Increment counts if at end of block/paragraph/textline.
    if (res_it->IsAtFinalElement(RIL_TEXTLINE, RIL_WORD)) lcnt++;
    if (res_it->IsAtFinalElement(RIL_PARA, RIL_WORD)) pcnt++;
    if (res_it->IsAtFinalElement(RIL_BLOCK, RIL_WORD)) bcnt++;

    do {
      tsv_str +=
          std::unique_ptr<const char[]>(res_it->GetUTF8Text(RIL_SYMBOL)).get();
      res_it->Next(RIL_SYMBOL);
    } while (!res_it->Empty(RIL_BLOCK) && !res_it->IsAtBeginningOf(RIL_WORD));
    tsv_str += "\n";  // end of row
    wcnt++;
  }

  char* ret = new char[tsv_str.length() + 1];
  strcpy(ret, tsv_str.string());
  delete res_it;
  return ret;
}

const int kNumbersPerBlob = 5;
const int kBytesPerNumber = 5;
const int kBytesPerBoxFileLine = (kBytesPerNumber + 1) * kNumbersPerBlob + 1;
const int kBytesPer64BitNumber = 20;
const int kMaxBytesPerLine = kNumbersPerBlob * (kBytesPer64BitNumber + 1) + 1 +
    UNICHAR_LEN;

char* TessBaseAPI::GetBoxText(int page_number) {
  if (tesseract_ == nullptr ||
      (!recognition_done_ && Recognize(nullptr) < 0))
    return nullptr;
  int blob_count;
  int utf8_length = TextLength(&blob_count);
  int total_length = blob_count * kBytesPerBoxFileLine + utf8_length +
      kMaxBytesPerLine;
  char* result = new char[total_length];
  result[0] = '\0';
  int output_length = 0;
  LTRResultIterator* it = GetLTRIterator();
  do {
    int left, top, right, bottom;
    if (it->BoundingBox(RIL_SYMBOL, &left, &top, &right, &bottom)) {
      const std::unique_ptr</*non-const*/ char[]> text(
          it->GetUTF8Text(RIL_SYMBOL));
      // Tesseract uses space for recognition failure. Fix to a reject
      // character, kTesseractReject so we don't create illegal box files.
      for (int i = 0; text[i] != '\0'; ++i) {
        if (text[i] == ' ')
          text[i] = kTesseractReject;
      }
      snprintf(result + output_length, total_length - output_length,
               "%s %d %d %d %d %d\n", text.get(), left, image_height_ - bottom,
               right, image_height_ - top, page_number);
      output_length += strlen(result + output_length);
      // Just in case...
      if (output_length + kMaxBytesPerLine > total_length)
        break;
    }
  } while (it->Next(RIL_SYMBOL));
  delete it;
  return result;
}

const int kUniChs[] = {
  0x20ac, 0x201c, 0x201d, 0x2018, 0x2019, 0x2022, 0x2014, 0
};
const int kLatinChs[] = {
  0x00a2, 0x0022, 0x0022, 0x0027, 0x0027, 0x00b7, 0x002d, 0
};

char* TessBaseAPI::GetUNLVText() {
  if (tesseract_ == nullptr ||
      (!recognition_done_ && Recognize(nullptr) < 0))
    return nullptr;
  bool tilde_crunch_written = false;
  bool last_char_was_newline = true;
  bool last_char_was_tilde = false;

  int total_length = TextLength(nullptr);
  PAGE_RES_IT   page_res_it(page_res_);
  char* result = new char[total_length];
  char* ptr = result;
  for (page_res_it.restart_page(); page_res_it.word () != nullptr;
       page_res_it.forward()) {
    WERD_RES *word = page_res_it.word();
    // Process the current word.
    if (word->unlv_crunch_mode != CR_NONE) {
      if (word->unlv_crunch_mode != CR_DELETE &&
          (!tilde_crunch_written ||
           (word->unlv_crunch_mode == CR_KEEP_SPACE &&
            word->word->space() > 0 &&
            !word->word->flag(W_FUZZY_NON) &&
            !word->word->flag(W_FUZZY_SP)))) {
        if (!word->word->flag(W_BOL) &&
            word->word->space() > 0 &&
            !word->word->flag(W_FUZZY_NON) &&
            !word->word->flag(W_FUZZY_SP)) {
          /* Write a space to separate from preceding good text */
          *ptr++ = ' ';
          last_char_was_tilde = false;
        }
        if (!last_char_was_tilde) {
          // Write a reject char.
          last_char_was_tilde = true;
          *ptr++ = kUNLVReject;
          tilde_crunch_written = true;
          last_char_was_newline = false;
        }
      }
    } else {
      // NORMAL PROCESSING of non tilde crunched words.
      tilde_crunch_written = false;
      tesseract_->set_unlv_suspects(word);
      const char* wordstr = word->best_choice->unichar_string().string();
      const STRING& lengths = word->best_choice->unichar_lengths();
      int length = lengths.length();
      int i = 0;
      int offset = 0;

      if (last_char_was_tilde &&
          word->word->space() == 0 && wordstr[offset] == ' ') {
        // Prevent adjacent tilde across words - we know that adjacent tildes
        // within words have been removed.
        // Skip the first character.
        offset = lengths[i++];
      }
      if (i < length && wordstr[offset] != 0) {
        if (!last_char_was_newline)
          *ptr++ = ' ';
        else
          last_char_was_newline = false;
        for (; i < length; offset += lengths[i++]) {
          if (wordstr[offset] == ' ' ||
              wordstr[offset] == kTesseractReject) {
            *ptr++ = kUNLVReject;
            last_char_was_tilde = true;
          } else {
            if (word->reject_map[i].rejected())
              *ptr++ = kUNLVSuspect;
            UNICHAR ch(wordstr + offset, lengths[i]);
            int uni_ch = ch.first_uni();
            for (int j = 0; kUniChs[j] != 0; ++j) {
              if (kUniChs[j] == uni_ch) {
                uni_ch = kLatinChs[j];
                break;
              }
            }
            if (uni_ch <= 0xff) {
              *ptr++ = static_cast<char>(uni_ch);
              last_char_was_tilde = false;
            } else {
              *ptr++ = kUNLVReject;
              last_char_was_tilde = true;
            }
          }
        }
      }
    }
    if (word->word->flag(W_EOL) && !last_char_was_newline) {
      /* Add a new line output */
      *ptr++ = '\n';
      tilde_crunch_written = false;
      last_char_was_newline = true;
      last_char_was_tilde = false;
    }
  }
  *ptr++ = '\n';
  *ptr = '\0';
  return result;
}

#ifndef DISABLED_LEGACY_ENGINE

bool TessBaseAPI::DetectOrientationScript(int* orient_deg, float* orient_conf,
                                          const char** script_name,
                                          float* script_conf) {
  OSResults osr;

  bool osd = DetectOS(&osr);
  if (!osd) {
    return false;
  }

  int orient_id = osr.best_result.orientation_id;
  int script_id = osr.get_best_script(orient_id);
  if (orient_conf) *orient_conf = osr.best_result.oconfidence;
  if (orient_deg) *orient_deg = orient_id * 90;  // convert quadrant to degrees

  if (script_name) {
    const char* script = osr.unicharset->get_script_from_script_id(script_id);

    *script_name = script;
  }

  if (script_conf) *script_conf = osr.best_result.sconfidence;

  return true;
}

char* TessBaseAPI::GetOsdText(int page_number) {
  int orient_deg;
  float orient_conf;
  const char* script_name;
  float script_conf;

  if (!DetectOrientationScript(&orient_deg, &orient_conf, &script_name,
                               &script_conf))
    return nullptr;

  // clockwise rotation needed to make the page upright
  int rotate = OrientationIdToValue(orient_deg / 90);

  const int kOsdBufsize = 255;
  char* osd_buf = new char[kOsdBufsize];
  snprintf(osd_buf, kOsdBufsize,
           "Page number: %d\n"
           "Orientation in degrees: %d\n"
           "Rotate: %d\n"
           "Orientation confidence: %.2f\n"
           "Script: %s\n"
           "Script confidence: %.2f\n",
           page_number, orient_deg, rotate, orient_conf, script_name,
           script_conf);

  return osd_buf;
}

#endif // ndef DISABLED_LEGACY_ENGINE

int TessBaseAPI::MeanTextConf() {
  int* conf = AllWordConfidences();
  if (!conf) return 0;
  int sum = 0;
  int *pt = conf;
  while (*pt >= 0) sum += *pt++;
  if (pt != conf) sum /= pt - conf;
  delete [] conf;
  return sum;
}

int* TessBaseAPI::AllWordConfidences() {
  if (tesseract_ == nullptr ||
      (!recognition_done_ && Recognize(nullptr) < 0))
    return nullptr;
  int n_word = 0;
  PAGE_RES_IT res_it(page_res_);
  for (res_it.restart_page(); res_it.word() != nullptr; res_it.forward())
    n_word++;

  int* conf = new int[n_word+1];
  n_word = 0;
  for (res_it.restart_page(); res_it.word() != nullptr; res_it.forward()) {
    WERD_RES *word = res_it.word();
    WERD_CHOICE* choice = word->best_choice;
    int w_conf = static_cast<int>(100 + 5 * choice->certainty());
                 // This is the eq for converting Tesseract confidence to 1..100
    if (w_conf < 0) w_conf = 0;
    if (w_conf > 100) w_conf = 100;
    conf[n_word++] = w_conf;
  }
  conf[n_word] = -1;
  return conf;
}

#ifndef DISABLED_LEGACY_ENGINE

bool TessBaseAPI::AdaptToWordStr(PageSegMode mode, const char* wordstr) {
  int debug = 0;
  GetIntVariable("applybox_debug", &debug);
  bool success = true;
  PageSegMode current_psm = GetPageSegMode();
  SetPageSegMode(mode);
  SetVariable("classify_enable_learning", "0");
  const std::unique_ptr<const char[]> text(GetUTF8Text());
  if (debug) {
    tprintf("Trying to adapt \"%s\" to \"%s\"\n", text.get(), wordstr);
  }
  if (text != nullptr) {
    PAGE_RES_IT it(page_res_);
    WERD_RES* word_res = it.word();
    if (word_res != nullptr) {
      word_res->word->set_text(wordstr);
      // Check to see if text matches wordstr.
      int w = 0;
      int t;
      for (t = 0; text[t] != '\0'; ++t) {
        if (text[t] == '\n' || text[t] == ' ')
          continue;
        while (wordstr[w] == ' ') ++w;
        if (text[t] != wordstr[w])
          break;
        ++w;
      }
      if (text[t] != '\0' || wordstr[w] != '\0') {
        // No match.
        delete page_res_;
        GenericVector<TBOX> boxes;
        page_res_ = tesseract_->SetupApplyBoxes(boxes, block_list_);
        tesseract_->ReSegmentByClassification(page_res_);
        tesseract_->TidyUp(page_res_);
        PAGE_RES_IT pr_it(page_res_);
        if (pr_it.word() == nullptr)
          success = false;
        else
          word_res = pr_it.word();
      } else {
        word_res->BestChoiceToCorrectText();
      }
      if (success) {
        tesseract_->EnableLearning = true;
        tesseract_->LearnWord(nullptr, word_res);
      }
    } else {
      success = false;
    }
  } else {
    success = false;
  }
  SetPageSegMode(current_psm);
  return success;
}
#endif  // ndef DISABLED_LEGACY_ENGINE

void TessBaseAPI::Clear() {
  if (thresholder_ != nullptr)
    thresholder_->Clear();
  ClearResults();
  if (tesseract_ != nullptr) SetInputImage(nullptr);
}

void TessBaseAPI::End() {
  Clear();
  delete thresholder_;
  thresholder_ = nullptr;
  delete page_res_;
  page_res_ = nullptr;
  delete block_list_;
  block_list_ = nullptr;
  if (paragraph_models_ != nullptr) {
    paragraph_models_->delete_data_pointers();
    delete paragraph_models_;
    paragraph_models_ = nullptr;
  }
  if (osd_tesseract_ == tesseract_) osd_tesseract_ = nullptr;
  delete tesseract_;
  tesseract_ = nullptr;
  delete osd_tesseract_;
  osd_tesseract_ = nullptr;
  delete equ_detect_;
  equ_detect_ = nullptr;
  delete input_file_;
  input_file_ = nullptr;
  delete output_file_;
  output_file_ = nullptr;
  delete datapath_;
  datapath_ = nullptr;
  delete language_;
  language_ = nullptr;
}

// Clear any library-level memory caches.
// There are a variety of expensive-to-load constant data structures (mostly
// language dictionaries) that are cached globally -- surviving the Init()
// and End() of individual TessBaseAPI's.  This function allows the clearing
// of these caches.
void TessBaseAPI::ClearPersistentCache() {
  Dict::GlobalDawgCache()->DeleteUnusedDawgs();
}

int TessBaseAPI::IsValidWord(const char *word) {
  return tesseract_->getDict().valid_word(word);
}
// Returns true if utf8_character is defined in the UniCharset.
bool TessBaseAPI::IsValidCharacter(const char *utf8_character) {
    return tesseract_->unicharset.contains_unichar(utf8_character);
}


// TODO(rays) Obsolete this function and replace with a more aptly named
// function that returns image coordinates rather than tesseract coordinates.
bool TessBaseAPI::GetTextDirection(int* out_offset, float* out_slope) {
  PageIterator* it = AnalyseLayout();
  if (it == nullptr) {
    return false;
  }
  int x1, x2, y1, y2;
  it->Baseline(RIL_TEXTLINE, &x1, &y1, &x2, &y2);
  // Calculate offset and slope (NOTE: Kind of ugly)
  if (x2 <= x1) x2 = x1 + 1;
  // Convert the point pair to slope/offset of the baseline (in image coords.)
  *out_slope = static_cast<float>(y2 - y1) / (x2 - x1);
  *out_offset = static_cast<int>(y1 - *out_slope * x1);
  // Get the y-coord of the baseline at the left and right edges of the
  // textline's bounding box.
  int left, top, right, bottom;
  if (!it->BoundingBox(RIL_TEXTLINE, &left, &top, &right, &bottom)) {
    delete it;
    return false;
  }
  int left_y = IntCastRounded(*out_slope * left + *out_offset);
  int right_y = IntCastRounded(*out_slope * right + *out_offset);
  // Shift the baseline down so it passes through the nearest bottom-corner
  // of the textline's bounding box. This is the difference between the y
  // at the lowest (max) edge of the box and the actual box bottom.
  *out_offset += bottom - std::max(left_y, right_y);
  // Switch back to bottom-up tesseract coordinates. Requires negation of
  // the slope and height - offset for the offset.
  *out_slope = -*out_slope;
  *out_offset = rect_height_ - *out_offset;
  delete it;

  return true;
}

void TessBaseAPI::SetDictFunc(DictFunc f) {
  if (tesseract_ != nullptr) {
    tesseract_->getDict().letter_is_okay_ = f;
  }
}

void TessBaseAPI::SetProbabilityInContextFunc(ProbabilityInContextFunc f) {
  if (tesseract_ != nullptr) {
    tesseract_->getDict().probability_in_context_ = f;
    // Set it for the sublangs too.
    int num_subs = tesseract_->num_sub_langs();
    for (int i = 0; i < num_subs; ++i) {
      tesseract_->get_sub_lang(i)->getDict().probability_in_context_ = f;
    }
  }
}

#ifndef DISABLED_LEGACY_ENGINE

void TessBaseAPI::SetFillLatticeFunc(FillLatticeFunc f) {
  if (tesseract_ != nullptr) tesseract_->fill_lattice_ = f;
}
#endif  // ndef DISABLED_LEGACY_ENGINE

bool TessBaseAPI::InternalSetImage() {
  if (tesseract_ == nullptr) {
    tprintf("Please call Init before attempting to set an image.\n");
    return false;
  }
  if (thresholder_ == nullptr)
    thresholder_ = new ImageThresholder;
  ClearResults();
  return true;
}

bool TessBaseAPI::Threshold(Pix** pix) {
  ASSERT_HOST(pix != nullptr);
  if (*pix != nullptr)
    pixDestroy(pix);
  // Zero resolution messes up the algorithms, so make sure it is credible.
  int user_dpi = 0;
  bool a = GetIntVariable("user_defined_dpi", &user_dpi);
  int y_res = thresholder_->GetScaledYResolution();
  if (user_dpi && (user_dpi < kMinCredibleResolution ||
      user_dpi > kMaxCredibleResolution)) {
    tprintf("Warning: User defined image dpi is outside of expected range "
            "(%d - %d)!\n",
            kMinCredibleResolution, kMaxCredibleResolution);
  }
  // Always use user defined dpi
  if (user_dpi) {
    thresholder_->SetSourceYResolution(user_dpi);
  } else if (y_res < kMinCredibleResolution ||
             y_res > kMaxCredibleResolution) {
    tprintf("Warning: Invalid resolution %d dpi. Using %d instead.\n",
            y_res, kMinCredibleResolution);
    thresholder_->SetSourceYResolution(kMinCredibleResolution);
  }
  PageSegMode pageseg_mode =
      static_cast<PageSegMode>(
          static_cast<int>(tesseract_->tessedit_pageseg_mode));
  if (!thresholder_->ThresholdToPix(pageseg_mode, pix)) return false;
  thresholder_->GetImageSizes(&rect_left_, &rect_top_,
                              &rect_width_, &rect_height_,
                              &image_width_, &image_height_);
  if (!thresholder_->IsBinary()) {
    tesseract_->set_pix_thresholds(thresholder_->GetPixRectThresholds());
    tesseract_->set_pix_grey(thresholder_->GetPixRectGrey());
  } else {
    tesseract_->set_pix_thresholds(nullptr);
    tesseract_->set_pix_grey(nullptr);
  }
  // Set the internal resolution that is used for layout parameters from the
  // estimated resolution, rather than the image resolution, which may be
  // fabricated, but we will use the image resolution, if there is one, to
  // report output point sizes.
  int estimated_res = ClipToRange(thresholder_->GetScaledEstimatedResolution(),
                                  kMinCredibleResolution,
                                  kMaxCredibleResolution);
  if (estimated_res != thresholder_->GetScaledEstimatedResolution()) {
    tprintf("Estimated internal resolution %d out of range! "
            "Corrected to %d.\n",
            thresholder_->GetScaledEstimatedResolution(), estimated_res);
  }
  tesseract_->set_source_resolution(estimated_res);
  SavePixForCrash(estimated_res, *pix);
  return true;
}

int TessBaseAPI::FindLines() {
  if (thresholder_ == nullptr || thresholder_->IsEmpty()) {
    tprintf("Please call SetImage before attempting recognition.\n");
    return -1;
  }
  if (recognition_done_)
    ClearResults();
  if (!block_list_->empty()) {
    return 0;
  }
  if (tesseract_ == nullptr) {
    tesseract_ = new Tesseract;
  #ifndef DISABLED_LEGACY_ENGINE
    tesseract_->InitAdaptiveClassifier(nullptr);
  #endif
  }
  if (tesseract_->pix_binary() == nullptr &&
      !Threshold(tesseract_->mutable_pix_binary())) {
    return -1;
  }

  tesseract_->PrepareForPageseg();

#ifndef DISABLED_LEGACY_ENGINE
  if (tesseract_->textord_equation_detect) {
    if (equ_detect_ == nullptr && datapath_ != nullptr) {
      equ_detect_ = new EquationDetect(datapath_->string(), nullptr);
    }
    if (equ_detect_ == nullptr) {
      tprintf("Warning: Could not set equation detector\n");
    } else {
      tesseract_->SetEquationDetect(equ_detect_);
    }
  }
#endif  // ndef DISABLED_LEGACY_ENGINE

  Tesseract* osd_tess = osd_tesseract_;
  OSResults osr;
  if (PSM_OSD_ENABLED(tesseract_->tessedit_pageseg_mode) &&
      osd_tess == nullptr) {
    if (strcmp(language_->string(), "osd") == 0) {
      osd_tess = tesseract_;
    } else {
      osd_tesseract_ = new Tesseract;
      TessdataManager mgr(reader_);
      if (datapath_ == nullptr) {
        tprintf("Warning: Auto orientation and script detection requested,"
                " but data path is undefined\n");
        delete osd_tesseract_;
        osd_tesseract_ = nullptr;
      } else if (osd_tesseract_->init_tesseract(datapath_->string(), nullptr,
                                                "osd", OEM_TESSERACT_ONLY,
                                                nullptr, 0, nullptr, nullptr,
                                                false, &mgr) == 0) {
        osd_tess = osd_tesseract_;
        osd_tesseract_->set_source_resolution(
            thresholder_->GetSourceYResolution());
      } else {
        tprintf("Warning: Auto orientation and script detection requested,"
                " but osd language failed to load\n");
        delete osd_tesseract_;
        osd_tesseract_ = nullptr;
      }
    }
  }

  if (tesseract_->SegmentPage(input_file_, block_list_, osd_tess, &osr) < 0)
    return -1;

  // If Devanagari is being recognized, we use different images for page seg
  // and for OCR.
  tesseract_->PrepareForTessOCR(block_list_, osd_tess, &osr);
  return 0;
}

void TessBaseAPI::ClearResults() {
  if (tesseract_ != nullptr) {
    tesseract_->Clear();
  }
  delete page_res_;
  page_res_ = nullptr;
  recognition_done_ = false;
  if (block_list_ == nullptr)
    block_list_ = new BLOCK_LIST;
  else
    block_list_->clear();
  if (paragraph_models_ != nullptr) {
    paragraph_models_->delete_data_pointers();
    delete paragraph_models_;
    paragraph_models_ = nullptr;
  }
  SavePixForCrash(0, nullptr);
}

int TessBaseAPI::TextLength(int* blob_count) {
  if (tesseract_ == nullptr || page_res_ == nullptr)
    return 0;

  PAGE_RES_IT   page_res_it(page_res_);
  int total_length = 2;
  int total_blobs = 0;
  // Iterate over the data structures to extract the recognition result.
  for (page_res_it.restart_page(); page_res_it.word () != nullptr;
       page_res_it.forward()) {
    WERD_RES *word = page_res_it.word();
    WERD_CHOICE* choice = word->best_choice;
    if (choice != nullptr) {
      total_blobs += choice->length() + 2;
      total_length += choice->unichar_string().length() + 2;
      for (int i = 0; i < word->reject_map.length(); ++i) {
        if (word->reject_map[i].rejected())
          ++total_length;
      }
    }
  }
  if (blob_count != nullptr)
    *blob_count = total_blobs;
  return total_length;
}

#ifndef DISABLED_LEGACY_ENGINE

bool TessBaseAPI::DetectOS(OSResults* osr) {
  if (tesseract_ == nullptr)
    return false;
  ClearResults();
  if (tesseract_->pix_binary() == nullptr &&
      !Threshold(tesseract_->mutable_pix_binary())) {
    return false;
  }

  if (input_file_ == nullptr)
    input_file_ = new STRING(kInputFile);
  return orientation_and_script_detection(*input_file_, osr, tesseract_) > 0;
}
#endif  // ndef DISABLED_LEGACY_ENGINE

void TessBaseAPI::set_min_orientation_margin(double margin) {
  tesseract_->min_orientation_margin.set_value(margin);
}

void TessBaseAPI::GetBlockTextOrientations(int** block_orientation,
                                           bool** vertical_writing) {
  delete[] *block_orientation;
  *block_orientation = nullptr;
  delete[] *vertical_writing;
  *vertical_writing = nullptr;
  BLOCK_IT block_it(block_list_);

  block_it.move_to_first();
  int num_blocks = 0;
  for (block_it.mark_cycle_pt(); !block_it.cycled_list(); block_it.forward()) {
    if (!block_it.data()->pdblk.poly_block()->IsText()) {
      continue;
    }
    ++num_blocks;
  }
  if (!num_blocks) {
    tprintf("WARNING: Found no blocks\n");
    return;
  }
  *block_orientation = new int[num_blocks];
  *vertical_writing = new bool[num_blocks];
  block_it.move_to_first();
  int i = 0;
  for (block_it.mark_cycle_pt(); !block_it.cycled_list();
       block_it.forward()) {
    if (!block_it.data()->pdblk.poly_block()->IsText()) {
      continue;
    }
    FCOORD re_rotation = block_it.data()->re_rotation();
    float re_theta = re_rotation.angle();
    FCOORD classify_rotation = block_it.data()->classify_rotation();
    float classify_theta = classify_rotation.angle();
    double rot_theta = - (re_theta - classify_theta) * 2.0 / M_PI;
    if (rot_theta < 0) rot_theta += 4;
    int num_rotations = static_cast<int>(rot_theta + 0.5);
    (*block_orientation)[i] = num_rotations;
    // The classify_rotation is non-zero only if the text has vertical
    // writing direction.
    (*vertical_writing)[i] = classify_rotation.y() != 0.0f;
    ++i;
  }
}


void TessBaseAPI::DetectParagraphs(bool after_text_recognition) {
  int debug_level = 0;
  GetIntVariable("paragraph_debug_level", &debug_level);
  if (paragraph_models_ == nullptr)
    paragraph_models_ = new GenericVector<ParagraphModel*>;
  MutableIterator *result_it = GetMutableIterator();
  do {  // Detect paragraphs for this block
    GenericVector<ParagraphModel *> models;
    ::tesseract::DetectParagraphs(debug_level, after_text_recognition,
                                  result_it, &models);
    *paragraph_models_ += models;
  } while (result_it->Next(RIL_BLOCK));
  delete result_it;
}

const char* TessBaseAPI::GetUnichar(int unichar_id) {
  return tesseract_->unicharset.id_to_unichar(unichar_id);
}

const Dawg *TessBaseAPI::GetDawg(int i) const {
  if (tesseract_ == nullptr || i >= NumDawgs()) return nullptr;
  return tesseract_->getDict().GetDawg(i);
}

int TessBaseAPI::NumDawgs() const {
  return tesseract_ == nullptr ? 0 : tesseract_->getDict().NumDawgs();
}

STRING HOcrEscape(const char* text) {
  STRING ret;
  const char *ptr;
  for (ptr = text; *ptr; ptr++) {
    switch (*ptr) {
      case '<': ret += "&lt;"; break;
      case '>': ret += "&gt;"; break;
      case '&': ret += "&amp;"; break;
      case '"': ret += "&quot;"; break;
      case '\'': ret += "&#39;"; break;
      default: ret += *ptr;
    }
  }
  return ret;
}


#ifndef DISABLED_LEGACY_ENGINE


// ____________________________________________________________________________
// Ocropus add-ons.

BLOCK_LIST* TessBaseAPI::FindLinesCreateBlockList() {
  ASSERT_HOST(FindLines() == 0);
  BLOCK_LIST* result = block_list_;
  block_list_ = nullptr;
  return result;
}

void TessBaseAPI::DeleteBlockList(BLOCK_LIST *block_list) {
  delete block_list;
}


ROW *TessBaseAPI::MakeTessOCRRow(float baseline,
                                 float xheight,
                                 float descender,
                                 float ascender) {
  int32_t xstarts[] = {-32000};
  double quad_coeffs[] = {0, 0, baseline};
  return new ROW(1,
                 xstarts,
                 quad_coeffs,
                 xheight,
                 ascender - (baseline + xheight),
                 descender - baseline,
                 0,
                 0);
}

TBLOB *TessBaseAPI::MakeTBLOB(Pix *pix) {
  int width = pixGetWidth(pix);
  int height = pixGetHeight(pix);
  BLOCK block("a character", TRUE, 0, 0, 0, 0, width, height);

  // Create C_BLOBs from the page
  extract_edges(pix, &block);

  // Merge all C_BLOBs
  C_BLOB_LIST *list = block.blob_list();
  C_BLOB_IT c_blob_it(list);
  if (c_blob_it.empty())
    return nullptr;
  // Move all the outlines to the first blob.
  C_OUTLINE_IT ol_it(c_blob_it.data()->out_list());
  for (c_blob_it.forward();
       !c_blob_it.at_first();
       c_blob_it.forward()) {
      C_BLOB *c_blob = c_blob_it.data();
      ol_it.add_list_after(c_blob->out_list());
  }
  // Convert the first blob to the output TBLOB.
  return TBLOB::PolygonalCopy(false, c_blob_it.data());
}

void TessBaseAPI::NormalizeTBLOB(TBLOB *tblob, ROW *row, bool numeric_mode) {
  TBOX box = tblob->bounding_box();
  float x_center = (box.left() + box.right()) / 2.0f;
  float baseline = row->base_line(x_center);
  float scale = kBlnXHeight / row->x_height();
  tblob->Normalize(nullptr, nullptr, nullptr, x_center, baseline, scale, scale,
                   0.0f, static_cast<float>(kBlnBaselineOffset), false, nullptr);
}

static TBLOB *make_tesseract_blob(float baseline, float xheight,
                                  float descender, float ascender,
                                  bool numeric_mode, Pix* pix) {
  TBLOB *tblob = TessBaseAPI::MakeTBLOB(pix);

  // Normalize TBLOB
  ROW *row =
      TessBaseAPI::MakeTessOCRRow(baseline, xheight, descender, ascender);
  TessBaseAPI::NormalizeTBLOB(tblob, row, numeric_mode);
  delete row;
  return tblob;
}

void TessBaseAPI::AdaptToCharacter(const char *unichar_repr,
                                   int length,
                                   float baseline,
                                   float xheight,
                                   float descender,
                                   float ascender) {
  UNICHAR_ID id = tesseract_->unicharset.unichar_to_id(unichar_repr, length);
  TBLOB *blob = make_tesseract_blob(baseline, xheight, descender, ascender,
                                    tesseract_->classify_bln_numeric_mode,
                                    tesseract_->pix_binary());
  float threshold;
  float best_rating = -100;


  // Classify to get a raw choice.
  BLOB_CHOICE_LIST choices;
  tesseract_->AdaptiveClassifier(blob, &choices);
  BLOB_CHOICE_IT choice_it;
  choice_it.set_to_list(&choices);
  for (choice_it.mark_cycle_pt(); !choice_it.cycled_list();
       choice_it.forward()) {
    if (choice_it.data()->rating() > best_rating) {
      best_rating = choice_it.data()->rating();
    }
  }

  threshold = tesseract_->matcher_good_threshold;

  if (blob->outlines)
    tesseract_->AdaptToChar(blob, id, kUnknownFontinfoId, threshold,
                            tesseract_->AdaptedTemplates);
  delete blob;
}


PAGE_RES* TessBaseAPI::RecognitionPass1(BLOCK_LIST* block_list) {
  PAGE_RES *page_res = new PAGE_RES(false, block_list,
                                    &(tesseract_->prev_word_best_choice_));
  tesseract_->recog_all_words(page_res, nullptr, nullptr, nullptr, 1);
  return page_res;
}

PAGE_RES* TessBaseAPI::RecognitionPass2(BLOCK_LIST* block_list,
                                        PAGE_RES* pass1_result) {
  if (!pass1_result)
    pass1_result = new PAGE_RES(false, block_list,
                                &(tesseract_->prev_word_best_choice_));
  tesseract_->recog_all_words(pass1_result, nullptr, nullptr, nullptr, 2);
  return pass1_result;
}

struct TESS_CHAR : ELIST_LINK {
  char *unicode_repr;
  int length;  // of unicode_repr
  float cost;
  TBOX box;

  TESS_CHAR(float _cost, const char *repr, int len = -1) : cost(_cost) {
    length = (len == -1 ? strlen(repr) : len);
    unicode_repr = new char[length + 1];
    strncpy(unicode_repr, repr, length);
  }

  TESS_CHAR()
    : unicode_repr(nullptr),
      length(0),
      cost(0.0f)
  {  // Satisfies ELISTIZE.
  }
  ~TESS_CHAR() {
    delete [] unicode_repr;
  }
};

ELISTIZEH(TESS_CHAR)
ELISTIZE(TESS_CHAR)

static void add_space(TESS_CHAR_IT* it) {
  TESS_CHAR *t = new TESS_CHAR(0, " ");
  it->add_after_then_move(t);
}


static float rating_to_cost(float rating) {
  rating = 100 + rating;
  // cuddled that to save from coverage profiler
  // (I have never seen ratings worse than -100,
  //  but the check won't hurt)
  if (rating < 0) rating = 0;
  return rating;
}

static void extract_result(TESS_CHAR_IT* out,
                           PAGE_RES* page_res) {
  PAGE_RES_IT page_res_it(page_res);
  int word_count = 0;
  while (page_res_it.word() != nullptr) {
    WERD_RES *word = page_res_it.word();
    const char *str = word->best_choice->unichar_string().string();
    const char *len = word->best_choice->unichar_lengths().string();
    TBOX real_rect = word->word->bounding_box();

    if (word_count)
      add_space(out);
    int n = strlen(len);
    for (int i = 0; i < n; i++) {
      TESS_CHAR *tc = new TESS_CHAR(rating_to_cost(word->best_choice->rating()),
                                    str, *len);
      tc->box = real_rect.intersection(word->box_word->BlobBox(i));
      out->add_after_then_move(tc);
       str += *len;
      len++;
    }
    page_res_it.forward();
    word_count++;
  }
}

int TessBaseAPI::TesseractExtractResult(char** text,
                                        int** lengths,
                                        float** costs,
                                        int** x0,
                                        int** y0,
                                        int** x1,
                                        int** y1,
                                        PAGE_RES* page_res) {
  TESS_CHAR_LIST tess_chars;
  TESS_CHAR_IT tess_chars_it(&tess_chars);
  extract_result(&tess_chars_it, page_res);
  tess_chars_it.move_to_first();
  int n = tess_chars.length();
  int text_len = 0;
  *lengths = new int[n];
  *costs = new float[n];
  *x0 = new int[n];
  *y0 = new int[n];
  *x1 = new int[n];
  *y1 = new int[n];
  int i = 0;
  for (tess_chars_it.mark_cycle_pt();
       !tess_chars_it.cycled_list();
       tess_chars_it.forward(), i++) {
    TESS_CHAR *tc = tess_chars_it.data();
    text_len += (*lengths)[i] = tc->length;
    (*costs)[i] = tc->cost;
    (*x0)[i] = tc->box.left();
    (*y0)[i] = tc->box.bottom();
    (*x1)[i] = tc->box.right();
    (*y1)[i] = tc->box.top();
  }
  char *p = *text = new char[text_len];

  tess_chars_it.move_to_first();
  for (tess_chars_it.mark_cycle_pt();
        !tess_chars_it.cycled_list();
       tess_chars_it.forward()) {
    TESS_CHAR *tc = tess_chars_it.data();
    strncpy(p, tc->unicode_repr, tc->length);
    p += tc->length;
  }
  return n;
}

// The resulting features are returned in int_features, which must be
// of size MAX_NUM_INT_FEATURES. The number of features is returned in
// num_features (or 0 if there was a failure).
// On return feature_outline_index is filled with an index of the outline
// corresponding to each feature in int_features.
// TODO(rays) Fix the caller to out outline_counts instead.
void TessBaseAPI::GetFeaturesForBlob(TBLOB* blob,
                                     INT_FEATURE_STRUCT* int_features,
                                     int* num_features,
                                     int* feature_outline_index) {
  GenericVector<int> outline_counts;
  GenericVector<INT_FEATURE_STRUCT> bl_features;
  GenericVector<INT_FEATURE_STRUCT> cn_features;
  INT_FX_RESULT_STRUCT fx_info;
  tesseract_->ExtractFeatures(*blob, false, &bl_features,
                              &cn_features, &fx_info, &outline_counts);
  if (cn_features.empty() || cn_features.size() > MAX_NUM_INT_FEATURES) {
    *num_features = 0;
    return;  // Feature extraction failed.
  }
  *num_features = cn_features.size();
  memcpy(int_features, &cn_features[0], *num_features * sizeof(cn_features[0]));
  // TODO(rays) Pass outline_counts back and simplify the calling code.
  if (feature_outline_index != nullptr) {
    int f = 0;
    for (int i = 0; i < outline_counts.size(); ++i) {
      while (f < outline_counts[i])
        feature_outline_index[f++] = i;
    }
  }
}

// This method returns the row to which a box of specified dimensions would
// belong. If no good match is found, it returns nullptr.
ROW* TessBaseAPI::FindRowForBox(BLOCK_LIST* blocks,
                                int left, int top, int right, int bottom) {
  TBOX box(left, bottom, right, top);
  BLOCK_IT b_it(blocks);
  for (b_it.mark_cycle_pt(); !b_it.cycled_list(); b_it.forward()) {
    BLOCK* block = b_it.data();
    if (!box.major_overlap(block->pdblk.bounding_box()))
      continue;
    ROW_IT r_it(block->row_list());
    for (r_it.mark_cycle_pt(); !r_it.cycled_list(); r_it.forward()) {
      ROW* row = r_it.data();
      if (!box.major_overlap(row->bounding_box()))
        continue;
      WERD_IT w_it(row->word_list());
      for (w_it.mark_cycle_pt(); !w_it.cycled_list(); w_it.forward()) {
        WERD* word = w_it.data();
        if (box.major_overlap(word->bounding_box()))
          return row;
      }
    }
  }
  return nullptr;
}

void TessBaseAPI::RunAdaptiveClassifier(TBLOB* blob,
                                        int num_max_matches,
                                        int* unichar_ids,
                                        float* ratings,
                                        int* num_matches_returned) {
  BLOB_CHOICE_LIST* choices = new BLOB_CHOICE_LIST;
  tesseract_->AdaptiveClassifier(blob, choices);
  BLOB_CHOICE_IT choices_it(choices);
  int& index = *num_matches_returned;
  index = 0;
  for (choices_it.mark_cycle_pt();
       !choices_it.cycled_list() && index < num_max_matches;
       choices_it.forward()) {
    BLOB_CHOICE* choice = choices_it.data();
    unichar_ids[index] = choice->unichar_id();
    ratings[index] = choice->rating();
    ++index;
  }
  *num_matches_returned = index;
  delete choices;
}
#endif  // ndef DISABLED_LEGACY_ENGINE

}  // namespace tesseract.