A contour analysis based technique to extract objects for MPEG-4

MPEG-4 is an emerging global standard for digital multimedia services. The core development of this video standard is a content based video data structure, which constitutes arbitrarily shaped video objects. The data compression is designed in terms of these video objects rather than frames. Thus, it is a key requirement in the development stages of MPEG-4 to explore possible means of object extraction from video frames. We propose a contour based technique to extract video objects for MPEG-4. The object contours are extracted by convolving a given frame with a Laplacian-of-Gaussian operator, followed by an edge detection process. The contours are later blocked and finally filled with the aid of a pixel based, parity check filling algorithm. Experimental results for several object extractions on 'Lena' and 'Peppers' images are included.