Using lifting wavelet to segment moving objects from videos

In this paper, we present an algorithm for segmenting moving objects from videos based on lifting wavelet transform. We firstly segment a video into scenes. Then several iteration of 1D lifting wavelet transforms are performed on each scene on time axis. Because wavelets have the ability of telling apart different frequency, we can extract the sections with high frequency, which represent the moving objects in a video. At last, through binarization and morphologic processing, the moving objects are completely segmented out. Experiments on videos show that our algorithm is very fast and feasible.