DVD Subtitles Technical Documentation

Overview

This project was done as a part of GSoC 2016. DVD subtitles extraction works by filtering out the subtitle frames from the video stream and obtaining the RLE encoded bitmap based subtitles, which are then provided to Tesseract for OCR recognition.

Dependencies

• Tesseract (OCR library by Google)
• Leptonica (C Image processing library)

The instructions for compilation along with the OCR can be found in the CCExtractor docs at docs/OCR.txt. General usage instructions can be found in the help screen of CCExtractor.

Code Structure

The DVD subtitle decoder is contained in the dvd_subtitle_decoder.c file and houses all required functions. The subtitles frames are selected from rest of the video and audio frames in general_loop.c and the process_spu() function receives only the subtitle packet in the buffer.

The structure DVD_Ctx has the context for the DVD subtitles while ctrl_seq has data of the control packet. The function process\_spu() checks for and creates a usable packet as some data might be spread over multiple packets.

Now we need the data in the control sequence which is handled by decode_packet(). Data from the control sequence including start and stop time, size of the subtitle, color, alpha and address of bitmap are extracted and stored in ctrl_seq.

Now we need to obtain the bitmap (via get_bitmap()). The bitmaps however are Run-Length Encoded (RLE) and are interlaced such that alternate lines follow each other. This is handled by providing first one half of the bitmap, followed by the other to rle_decode to account for the interlacing as well. At the end we get a clean bitmap which is then sent to the OCR invoked in write_dvd_sub().