All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Modules Pages
Go to the documentation of this file.
1 // File: imagefind.h
3 // Description: Class to find image and drawing regions in an image
4 // and create a corresponding list of empty blobs.
5 // Author: Ray Smith
6 // Created: Fri Aug 01 10:50:01 PDT 2008
7 //
8 // (C) Copyright 2008, Google Inc.
9 // Licensed under the Apache License, Version 2.0 (the "License");
10 // you may not use this file except in compliance with the License.
11 // You may obtain a copy of the License at
12 //
13 // Unless required by applicable law or agreed to in writing, software
14 // distributed under the License is distributed on an "AS IS" BASIS,
15 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 // See the License for the specific language governing permissions and
17 // limitations under the License.
18 //
24 #include "host.h"
26 struct Boxa;
27 struct Pix;
28 struct Pixa;
29 class TBOX;
30 class FCOORD;
31 class TO_BLOCK;
34 namespace tesseract {
36 class ColPartitionGrid;
37 class ColPartition_LIST;
38 class TabFind;
40 // The ImageFind class is a simple static function wrapper class that
41 // exposes the FindImages function and some useful helper functions.
42 class ImageFind {
43  public:
44  // Finds image regions within the BINARY source pix (page image) and returns
45  // the image regions as a mask image.
46  // The returned pix may be NULL, meaning no images found.
47  // If not NULL, it must be PixDestroyed by the caller.
48  static Pix* FindImages(Pix* pix);
50  // Generates a Boxa, Pixa pair from the input binary (image mask) pix,
51  // analgous to pixConnComp, except that connected components which are nearly
52  // rectangular are replaced with solid rectangles.
53  // The returned boxa, pixa may be NULL, meaning no images found.
54  // If not NULL, they must be destroyed by the caller.
55  // Resolution of pix should match the source image (Tesseract::pix_binary_)
56  // so the output coordinate systems match.
57  static void ConnCompAndRectangularize(Pix* pix, Boxa** boxa, Pixa** pixa);
59  // Returns true if there is a rectangle in the source pix, such that all
60  // pixel rows and column slices outside of it have less than
61  // min_fraction of the pixels black, and within max_skew_gradient fraction
62  // of the pixels on the inside, there are at least max_fraction of the
63  // pixels black. In other words, the inside of the rectangle looks roughly
64  // rectangular, and the outside of it looks like extra bits.
65  // On return, the rectangle is defined by x_start, y_start, x_end and y_end.
66  // Note: the algorithm is iterative, allowing it to slice off pixels from
67  // one edge, allowing it to then slice off more pixels from another edge.
68  static bool pixNearlyRectangular(Pix* pix,
69  double min_fraction, double max_fraction,
70  double max_skew_gradient,
71  int* x_start, int* y_start,
72  int* x_end, int* y_end);
74  // Given an input pix, and a bounding rectangle, the sides of the rectangle
75  // are shrunk inwards until they bound any black pixels found within the
76  // original rectangle. Returns false if the rectangle contains no black
77  // pixels at all.
78  static bool BoundsWithinRect(Pix* pix, int* x_start, int* y_start,
79  int* x_end, int* y_end);
81  // Given a point in 3-D (RGB) space, returns the squared Euclidean distance
82  // of the point from the given line, defined by a pair of points in the 3-D
83  // (RGB) space, line1 and line2.
84  static double ColorDistanceFromLine(const uinT8* line1, const uinT8* line2,
85  const uinT8* point);
87  // Returns the leptonica combined code for the given RGB triplet.
88  static uinT32 ComposeRGB(uinT32 r, uinT32 g, uinT32 b);
90  // Returns the input value clipped to a uinT8.
91  static uinT8 ClipToByte(double pixel);
93  // Computes the light and dark extremes of color in the given rectangle of
94  // the given pix, which is factor smaller than the coordinate system in rect.
95  // The light and dark points are taken to be the upper and lower 8th-ile of
96  // the most deviant of R, G and B. The value of the other 2 channels are
97  // computed by linear fit against the most deviant.
98  // The colors of the two point are returned in color1 and color2, with the
99  // alpha channel set to a scaled mean rms of the fits.
100  // If color_map1 is not null then it and color_map2 get rect pasted in them
101  // with the two calculated colors, and rms map gets a pasted rect of the rms.
102  // color_map1, color_map2 and rms_map are assumed to be the same scale as pix.
103  static void ComputeRectangleColors(const TBOX& rect, Pix* pix, int factor,
104  Pix* color_map1, Pix* color_map2,
105  Pix* rms_map,
106  uinT8* color1, uinT8* color2);
108  // Returns true if there are no black pixels in between the boxes.
109  // The im_box must represent the bounding box of the pix in tesseract
110  // coordinates, which may be negative, due to rotations to make the textlines
111  // horizontal. The boxes are rotated by rotation, which should undo such
112  // rotations, before mapping them onto the pix.
113  static bool BlankImageInBetween(const TBOX& box1, const TBOX& box2,
114  const TBOX& im_box, const FCOORD& rotation,
115  Pix* pix);
117  // Returns the number of pixels in box in the pix.
118  // The im_box must represent the bounding box of the pix in tesseract
119  // coordinates, which may be negative, due to rotations to make the textlines
120  // horizontal. The boxes are rotated by rotation, which should undo such
121  // rotations, before mapping them onto the pix.
122  static int CountPixelsInRotatedBox(TBOX box, const TBOX& im_box,
123  const FCOORD& rotation, Pix* pix);
126  // Locates all the image partitions in the part_grid, that were found by a
127  // previous call to FindImagePartitions, marks them in the image_mask,
128  // removes them from the grid, and deletes them. This makes it possble to
129  // call FindImagePartitions again to produce less broken-up and less
130  // overlapping image partitions.
131  // rerotation specifies how to rotate the partition coords to match
132  // the image_mask, since this function is used after orientation correction.
133  static void TransferImagePartsToImageMask(const FCOORD& rerotation,
134  ColPartitionGrid* part_grid,
135  Pix* image_mask);
137  // Runs a CC analysis on the image_pix mask image, and creates
138  // image partitions from them, cutting out strong text, and merging with
139  // nearby image regions such that they don't interfere with text.
140  // Rotation and rerotation specify how to rotate image coords to match
141  // the blob and partition coords and back again.
142  // The input/output part_grid owns all the created partitions, and
143  // the partitions own all the fake blobs that belong in the partitions.
144  // Since the other blobs in the other partitions will be owned by the block,
145  // ColPartitionGrid::ReTypeBlobs must be called afterwards to fix this
146  // situation and collect the image blobs.
147  static void FindImagePartitions(Pix* image_pix,
148  const FCOORD& rotation,
149  const FCOORD& rerotation,
150  TO_BLOCK* block,
151  TabFind* tab_grid,
152  ColPartitionGrid* part_grid,
153  ColPartition_LIST* big_parts);
154 };
156 } // namespace tesseract.
static Pix * FindImages(Pix *pix)
Definition: imagefind.cpp:65
static int CountPixelsInRotatedBox(TBOX box, const TBOX &im_box, const FCOORD &rotation, Pix *pix)
Definition: imagefind.cpp:573
static bool BoundsWithinRect(Pix *pix, int *x_start, int *y_start, int *x_end, int *y_end)
Definition: imagefind.cpp:308
static void ComputeRectangleColors(const TBOX &rect, Pix *pix, int factor, Pix *color_map1, Pix *color_map2, Pix *rms_map, uinT8 *color1, uinT8 *color2)
Definition: imagefind.cpp:390
static double ColorDistanceFromLine(const uinT8 *line1, const uinT8 *line2, const uinT8 *point)
Definition: imagefind.cpp:331
unsigned int uinT32
Definition: host.h:103
static void FindImagePartitions(Pix *image_pix, const FCOORD &rotation, const FCOORD &rerotation, TO_BLOCK *block, TabFind *tab_grid, ColPartitionGrid *part_grid, ColPartition_LIST *big_parts)
Definition: imagefind.cpp:1274
static bool pixNearlyRectangular(Pix *pix, double min_fraction, double max_fraction, double max_skew_gradient, int *x_start, int *y_start, int *x_end, int *y_end)
Definition: imagefind.cpp:242
Definition: rect.h:30
static uinT32 ComposeRGB(uinT32 r, uinT32 g, uinT32 b)
Definition: imagefind.cpp:365
static void ConnCompAndRectangularize(Pix *pix, Boxa **boxa, Pixa **pixa)
Definition: imagefind.cpp:133
static uinT8 ClipToByte(double pixel)
Definition: imagefind.cpp:372
static bool BlankImageInBetween(const TBOX &box1, const TBOX &box2, const TBOX &im_box, const FCOORD &rotation, Pix *pix)
Definition: imagefind.cpp:552
static void TransferImagePartsToImageMask(const FCOORD &rerotation, ColPartitionGrid *part_grid, Pix *image_mask)
Definition: imagefind.cpp:1221
Definition: points.h:189
unsigned char uinT8
Definition: host.h:99