| In this work, morphological image segmentation tools are applied to compressed video data. Included in the developed morphological tool set are algorithms that perform leveling, flat-zone and marker detection, and image foresting transformations. These algorithms have been written in C by the author, tested on single frames, and then incorporated into an MPEG decoder for video processing. When compared to working directly with uncompressed data, working in the compressed domain drastically reduces processing time and the number of regions that require further processing to achieve a meaningful segmentation of the video and the objects that compose it. Results are obtained without the use of a priori information regarding the video content and without human interaction. Post-processed data may be encoded as video objects in the MPEG-4 compression scheme. Although many optimizations are still possible, including parallel implementations, dedicated hardware schemes, and code reorganization, an average Intra-frame processing rate that exceeds fifteen frames-per-second on a Pentium III, 700 MHz PC has been achieved for 352 x 240 pixel images. |
