Real-time foreground-background segmentation based on improved codebook model

摘要

An improved codebook model is presented to efficiently segment foreground from dynamic scenes in the paper. The model divides video frames into blocks and takes into account the spatial relations between pixels and the adjacent pixels in the blocks rather than seen the pixels independent. It uses the block means instead of pixel wises to construct codebooks. After comparison the brightness distortion and lightness distortion of the incoming pixels with codebook in HSV (hue, saturation, value) color space, foreground can be determined and segmented from background. Then using the block mean updates the codebook if the number of the pixels considered as foreground in the block is over half the block size. Experiments results show that the improved algorithm is efficient and can be used to real-time foreground-background segmentation.