Invert subtitle image before the OCR

It makes it work way better on my side, especially
with subtitles containing multiple lines.

Author: MatMaul

src/vobsub2srt.c++

@@ -121,6 +121,16 @@ TessBaseAPI* init_tesseract(std::string tesseract_data_path, char const *tess_la
   return tess_base_api;
 }
 
+void invert_image(unsigned width, unsigned height,
+                  unsigned stride, unsigned char *image) {
+  for (unsigned y=0; y < height; y++) {
+    for (unsigned x=0; x < width; x++) {
+      unsigned index = y*stride + x;
+      image[index] = 255 - image[index];
+    }
+  }
+}
+
 void do_ocr(ocr_thread_t *ocr_thread, vector<sub_text_t> *conv_subs, mutex *mut,
             unsigned counter, unsigned width, unsigned height, unsigned stride,
             unsigned char *image_cpy, unsigned start_pts, unsigned end_pts, bool verb) {
@@ -153,6 +163,7 @@ void do_ocr(ocr_thread_t *ocr_thread, vector<sub_text_t> *conv_subs, mutex *mut,
 
 int main(int argc, char **argv) {
   bool dump_images = false;
+  bool disable_invert = false;
   bool verb = false;
   bool list_languages = false;
   std::string ifo_file;
@@ -186,6 +197,7 @@ int main(int argc, char **argv) {
       add_option("min-width", min_width, "Minimum width in pixels to consider a subpicture for OCR (Default: 9)").
       add_option("min-height", min_height, "Minimum height in pixels to consider a subpicture for OCR (Default: 1)").
       add_option("max-threads", max_threads, "Maximum number of threads to use to do the OCR, use 0 to autodetect the number of cores (Default: 0)").
+      add_option("disable-invert", disable_invert, "By default the image will be inverted before the OCR because it works better with black on white background").
       add_unnamed(subname, "subname", "name of the subtitle files WITHOUT .idx/.sub ending! (REQUIRED)");
     if(not opts.parse_cmd(argc, argv) or subname.empty()) {
       return 1;
@@ -342,6 +354,9 @@ int main(int argc, char **argv) {
       unsigned char *image_cpy = (unsigned char *)malloc(image_size);
       memcpy(image_cpy, image, image_size);
 
+      if (!disable_invert)
+        invert_image(width, height, stride, image_cpy);
+
       ocr_thread->done = false;
       ocr_thread->t = new thread(do_ocr, ocr_thread, &conv_subs, &mut, sub_counter, width, height, stride, image_cpy, start_pts, end_pts, verb);