An Efficient Video Objects Segmentation Algorithm

Video object segmentation is the most important for content-based multimedia applications. A real-time algorithm for fully automatic video objects segmentation based on the color consistency of human vision system (CCHVS) was proposed. The perceptually uniform Munsell color space was employed. Firstly, the color images were quantized not more than 256 colors to reduce the processing time. Then the relation between the NBS colors distance and human vision system was used to detect motion. The static information was accumulated to construct a reliable background image. Finally, the moving regions were segmented automatically by comparing the current frame with the constructed background image. In order to meet the real-time requirement, no computationally intensive operations were included, and the color quantization and frame difference techniques were improved. Experimental results show that a processing speed of 25 CIF frame per second can be achieved on a personal computer, accurate and consistent results can be obtained from complex scenes