A low bit rate video coder using content-based automatic segmentation and shape-adaptive DCT

摘要

It has been the endeavour of the processing community to develop content-based segmentation of video. That is, to separate out all the objects in the video into distinct video object planes (VOPs) and generate a layered bit-stream wherein each bit-stream corresponds to an individual object plane in the video. This work aspires towards the above goals. It is in synchrony with the targets of MPEG-4. We propose a general and robust algorithm for segmentation. Our algorithm does not assume any prior modelling of the objects being segmented, further, the bit rate generated is low as compression is achieved at two levels. Firstly, only the regions of the image that change across frames are coded and transmitted. Secondly, the extracted portion of the image is compressed using a flexible variation of DCT coding called shape-adaptive DCT. Typically, in video-conferencing applications wherein the goal is to minimise the transmission bandwidth, this algorithm produces very desirable results.