baseapi.cpp

来自「一个google的OCR源码」· C++ 代码 · 共 1,123 行 · 第 1/3 页

/**********************************************************************
 * File:        baseapi.cpp
 * Description: Simple API for calling tesseract.
 * Author:      Ray Smith
 * Created:     Fri Oct 06 15:35:01 PDT 2006
 *
 * (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 "baseapi.h"


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

#ifdef HAVE_LIBLEPT
// Include leptonica library only if autoconf (or makefile etc) tell us to.
#include "allheaders.h"
#endif

#include "tessedit.h"
#include "ocrclass.h"
#include "pageres.h"
#include "tessvars.h"
#include "control.h"
#include "applybox.h"
#include "pgedit.h"
#include "varabled.h"
#include "variables.h"
#include "output.h"
#include "globals.h"
#include "adaptmatch.h"
#include "edgblob.h"
#include "tessbox.h"
#include "tordvars.h"
#include "imgs.h"
#include "makerow.h"
#include "tstruct.h"
#include "tessout.h"
#include "tface.h"
#include "permute.h"

BOOL_VAR(tessedit_resegment_from_boxes, FALSE,
         "Take segmentation and labeling from box file");
BOOL_VAR(tessedit_train_from_boxes, FALSE,
         "Generate training data from boxed chars");

// Minimum sensible image size to be worth running tesseract.
const int kMinRectSize = 10;

static STRING input_file = "noname.tif";

// Set the value of an internal "variable" (of either old or new types).
// Supply the name of the variable and the value as a string, just as
// you would in a config file.
// Returns false if the name lookup failed.
bool TessBaseAPI::SetVariable(const char* variable, const char* value) {
  if (set_new_style_variable(variable, value))
    return true;
  return set_old_style_variable(variable, value);
}

void TessBaseAPI::SimpleInit(const char* datapath,
                             const char* language,
                             bool numeric_mode) {
  InitWithLanguage(datapath, NULL, language, NULL, numeric_mode, 0, NULL);
}

// Start tesseract.
// The datapath must be the name of the data directory or some other file
// in which the data directory resides (for instance argv[0].)
// The configfile is the name of a file in the tessconfigs directory
// (eg batch) or NULL to run on defaults.
// Outputbase may also be NULL, and is the basename of various output files.
// If the output of any of these files is enabled, then a name nmust be given.
// If numeric_mode is true, only possible digits and roman numbers are
// returned. Returns 0 if successful. Crashes if not.
// The argc and argv may be 0 and NULL respectively. They are used for
// providing config files for debug/display purposes.
// TODO(rays) get the facts straight. Is it OK to call
// it more than once? Make it properly check for errors and return them.
int TessBaseAPI::Init(const char* datapath, const char* outputbase,
                              const char* configfile, bool numeric_mode,
                              int argc, char* argv[]) {
  return InitWithLanguage(datapath, outputbase, NULL, configfile,
                          numeric_mode, argc, argv);
}

// Start tesseract.
// Similar to Init() except that it is possible to specify the language.
// Language is the code of the language for which the data will be loaded.
// (Codes follow ISO 639-3.) If it is NULL, english (eng) will be loaded.
int TessBaseAPI::InitWithLanguage(const char* datapath, const char* outputbase,
                              const char* language, const char* configfile,
                              bool numeric_mode, int argc, char* argv[]) {
  int result = init_tesseract(datapath, outputbase, language,
      configfile, argc, argv);

  bln_numericmode.set_value(numeric_mode);
  return result;
}

// Init the lang model component of Tesseract
int TessBaseAPI::InitLangMod(const char* datapath, const char* outputbase,
                                const char* language, const char* configfile,
                                bool numeric_mode, int argc, char* argv[]) {
  return init_tesseract_lm(datapath, outputbase, language,
      configfile, argc, argv);
}

// Set the name of the input file. Needed only for training and
// loading a UNLV zone file.
void TessBaseAPI::SetInputName(const char* name) {
  input_file = name;
}

// Recognize a rectangle from an image and return the result as a string.
// May be called many times for a single Init.
// Currently has no error checking.
// Greyscale of 8 and color of 24 or 32 bits per pixel may be given.
// Palette color images will not work properly and must be converted to
// 24 bit.
// Binary images of 1 bit per pixel may also be given but they must be
// byte packed with the MSB of the first byte being the first pixel, and a
// one pixel is WHITE. For binary images set bytes_per_pixel=0.
// The recognized text is returned as a char* which (in future will be coded
// as UTF8 and) must be freed with the delete [] operator.
char* TessBaseAPI::TesseractRect(const unsigned char* imagedata,
                                 int bytes_per_pixel,
                                 int bytes_per_line,
                                 int left, int top,
                                 int width, int height) {
  if (width < kMinRectSize || height < kMinRectSize)
    return NULL;  // Nothing worth doing.

  // Copy/Threshold the image to the tesseract global page_image.
  CopyImageToTesseract(imagedata, bytes_per_pixel, bytes_per_line,
                       left, top, width, height);

  return RecognizeToString();
}

// As TesseractRect but produces a box file as output.
char* TessBaseAPI::TesseractRectBoxes(const unsigned char* imagedata,
                                      int bytes_per_pixel,
                                      int bytes_per_line,
                                      int left, int top,
                                      int width, int height,
                                      int imageheight) {
  if (width < kMinRectSize || height < kMinRectSize)
  return NULL;  // Nothing worth doing.

  // Copy/Threshold the image to the tesseract global page_image.
  CopyImageToTesseract(imagedata, bytes_per_pixel, bytes_per_line,
                       left, top, width, height);

  BLOCK_LIST    block_list;

  FindLines(&block_list);

  // Now run the main recognition.
  PAGE_RES* page_res = Recognize(&block_list, NULL);

  return TesseractToBoxText(page_res, left, imageheight - (top + height));
}

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

  // Copy/Threshold the image to the tesseract global page_image.
  CopyImageToTesseract(imagedata, bytes_per_pixel, bytes_per_line,
                       left, top, width, height);

  BLOCK_LIST    block_list;

  FindLines(&block_list);

  // Now run the main recognition.
  PAGE_RES* page_res = Recognize(&block_list, NULL);

  return TesseractToUNLV(page_res);
}

// Call between pages or documents etc to free up memory and forget
// adaptive data.
void TessBaseAPI::ClearAdaptiveClassifier() {
  ResetAdaptiveClassifier();
}

// Close down tesseract and free up memory.
void TessBaseAPI::End() {
  ResetAdaptiveClassifier();
  end_tesseract();
}

// Dump the internal binary image to a PGM file.
void TessBaseAPI::DumpPGM(const char* filename) {
  IMAGELINE line;
  line.init(page_image.get_xsize());
  FILE *fp = fopen(filename, "w");
  fprintf(fp, "P5 " INT32FORMAT " " INT32FORMAT " 255\n", page_image.get_xsize(),
          page_image.get_ysize());
  for (int j = page_image.get_ysize()-1; j >= 0 ; --j) {
    page_image.get_line(0, j, page_image.get_xsize(), &line, 0);
    for (int i = 0; i < page_image.get_xsize(); ++i) {
      uinT8 b = line.pixels[i] ? 255 : 0;
      fwrite(&b, 1, 1, fp);
    }
  }
  fclose(fp);
}

#ifdef HAVE_LIBLEPT
// ONLY available if you have Leptonica installed.
// Get a copy of the thresholded global image from Tesseract.
Pix* TessBaseAPI::GetTesseractImage() {
  return page_image.ToPix();
}
#endif  // HAVE_LIBLEPT

// Copy the given image rectangle to Tesseract, with adaptive thresholding
// if the image is not already binary.
void TessBaseAPI::CopyImageToTesseract(const unsigned char* imagedata,
                                       int bytes_per_pixel,
                                       int bytes_per_line,
                                       int left, int top,
                                       int width, int height) {
  if (bytes_per_pixel > 0) {
    // Threshold grey or color.
    int* thresholds = new int[bytes_per_pixel];
    int* hi_values = new int[bytes_per_pixel];

    // Compute the thresholds.
    OtsuThreshold(imagedata, bytes_per_pixel, bytes_per_line,
                  left, top, left + width, top + height,
                  thresholds, hi_values);

    // Threshold the image to the tesseract global page_image.
    ThresholdRect(imagedata, bytes_per_pixel, bytes_per_line,
                  left, top, width, height,
                  thresholds, hi_values);
    delete [] thresholds;
    delete [] hi_values;
  } else {
    CopyBinaryRect(imagedata, bytes_per_line, left, top, width, height);
  }
}

// Compute the Otsu threshold(s) for the given image rectangle, making one
// for each channel. Each channel is always one byte per pixel.
// Returns an array of threshold values and an array of hi_values, such
// that a pixel value >threshold[channel] is considered foreground if
// hi_values[channel] is 0 or background if 1. A hi_value of -1 indicates
// that there is no apparent foreground. At least one hi_value will not be -1.
// thresholds and hi_values are assumed to be of bytes_per_pixel size.
void TessBaseAPI::OtsuThreshold(const unsigned char* imagedata,
                                int bytes_per_pixel,
                                int bytes_per_line,
                                int left, int top, int right, int bottom,
                                int* thresholds,
                                int* hi_values) {
  // Of all channels with no good hi_value, keep the best so we can always
  // produce at least one answer.
  int best_hi_value = 0;
  int best_hi_index = 0;
  bool any_good_hivalue = false;
  double best_hi_dist = 0.0;

  for (int ch = 0; ch < bytes_per_pixel; ++ch) {
    thresholds[ch] = 0;
    hi_values[ch] = -1;
    // Compute the histogram of the image rectangle.
    int histogram[256];
    HistogramRect(imagedata + ch, bytes_per_pixel, bytes_per_line,
                  left, top, right, bottom, histogram);
    int H;
    int best_omega_0;
    int best_t = OtsuStats(histogram, &H, &best_omega_0);
    if (best_omega_0 == 0 || best_omega_0 == H) {
       // This channel is empty.
       continue;
     }
    // To be a convincing foreground we must have a small fraction of H
    // or to be a convincing background we must have a large fraction of H.
    // In between we assume this channel contains no thresholding information.
    int hi_value = best_omega_0 < H * 0.5;
    thresholds[ch] = best_t;
    if (best_omega_0 > H * 0.75) {
      any_good_hivalue = true;
      hi_values[ch] = 0;
    }
    else if (best_omega_0 < H * 0.25) {
      any_good_hivalue = true;
      hi_values[ch] = 1;
    }
    else {
      // In case all channels are like this, keep the best of the bad lot.
      double hi_dist = hi_value ? (H - best_omega_0) : best_omega_0;
      if (hi_dist > best_hi_dist) {
        best_hi_dist = hi_dist;
        best_hi_value = hi_value;
        best_hi_index = ch;
      }
    }
  }
  if (!any_good_hivalue) {
    // Use the best of the ones that were not good enough.
    hi_values[best_hi_index] = best_hi_value;
  }
}

// Compute the histogram for the given image rectangle, and the given
// channel. (Channel pointed to by imagedata.) Each channel is always
// one byte per pixel.
// Bytes per pixel is used to skip channels not being
// counted with this call in a multi-channel (pixel-major) image.
// Histogram is always a 256 element array to count occurrences of
// each pixel value.
void TessBaseAPI::HistogramRect(const unsigned char* imagedata,
                                int bytes_per_pixel,
                                int bytes_per_line,
                                int left, int top, int right, int bottom,
                                int* histogram) {
  int width = right - left;
  memset(histogram, 0, sizeof(*histogram) * 256);
  const unsigned char* pixels = imagedata +
                                top*bytes_per_line +
                                left*bytes_per_pixel;
  for (int y = top; y < bottom; ++y) {
    for (int x = 0; x < width; ++x) {
      ++histogram[pixels[x * bytes_per_pixel]];
    }
    pixels += bytes_per_line;
  }
}

// Compute the Otsu threshold(s) for the given histogram.
// Also returns H = total count in histogram, and
// omega0 = count of histogram below threshold.
int TessBaseAPI::OtsuStats(const int* histogram,
                           int* H_out,
                           int* omega0_out) {
  int H = 0;
  double mu_T = 0.0;
  for (int i = 0; i < 256; ++i) {
    H += histogram[i];
    mu_T += i * histogram[i];
  }

  // Now maximize sig_sq_B over t.
  // http://www.ctie.monash.edu.au/hargreave/Cornall_Terry_328.pdf
  int best_t = -1;
  int omega_0, omega_1;
  int best_omega_0 = 0;
  double best_sig_sq_B = 0.0;
  double mu_0, mu_1, mu_t;
  omega_0 = 0;
  mu_t = 0.0;