DWT based hierarchical video segmentation

To provide multimedia applications with new functionalities, such as content-based interactivity and scalability, the new video coding standard, MPEG-4 relies on content-based representation. This means a prior decomposition of a video sequence into semantically meaningful, physical objects. We develop a fast yet robust color segmentation algorithm based on DWT and k-means clustering. After spatial segmentation, a dense optical flow field is estimated for the whole image, using which a motion parameter model is estimated for each region using a least squares technique. Adjacent regions with a coherent motion are then merged together to form moving objects. This leads to segmented video