video data in opencv

Question

Can anybody please tell me in what format the video data is saved in opencv, after decoding it by ffmpeg. I mean if it is saved as a matrix then is it 8bit/pixel kind of thing? If anybody can provide some documents or link explaining in details will be a great help. I also tried to google but couldn't find something related.

Thanks.

Answer 1

The definitive reference to examine is probably the code itself! Look at the function CvCapture_FFMPEG::open in the file modules/highgui/src/cap_ffmpeg_impl.hpp. This method is called by, for example, cvCaptureFromAVI() and cvCaptureFromFile(). The code excerpted below clearly gets the width and height from the file, but it ignores the input bit depth and always uses a packed 8:8:8 bit RGB representation of the video in memory (as indicated by the PIX_FMT_BGR24 constant):

#if LIBAVFORMAT_BUILD >= CALC_FFMPEG_VERSION(52, 111, 0)
    int err = avformat_open_input(&ic, _filename, NULL, NULL);
#else
    int err = av_open_input_file(&ic, _filename, NULL, 0, NULL);
#endif

    .......    

#if LIBAVFORMAT_BUILD >= CALC_FFMPEG_VERSION(53, 3, 0)
    avformat_find_stream_info(ic, NULL);
#else
    av_find_stream_info(ic);
#endif

    .......    

    avpicture_fill( (AVPicture*)&rgb_picture, rgb_picture.data[0],
                    PIX_FMT_BGR24, enc->width, enc->height );

Further evidence that this is the case is provided by the following code in CvCapture_FFMPEG::retrieveFrame:

img_convert_ctx = sws_getCachedContext(
                NULL,
                video_st->codec->width, video_st->codec->height,
                video_st->codec->pix_fmt,
                video_st->codec->width, video_st->codec->height,
                PIX_FMT_BGR24,
                SWS_BICUBIC,
                NULL, NULL, NULL
                );

Looking at the ffmpeg documentation, the PIX_FMT_BGR24 in this call specifies the destination pixel format, i.e., what will be handed off to opencv.