In this paper,a novel object-orientedhierarchicalvideosegmentation and representationalgorithm is proposed based on four-component video model,where the local variance contrast and the frame difference contrast ar...
详细信息
In this paper,a novel object-orientedhierarchicalvideosegmentation and representationalgorithm is proposed based on four-component video model,where the local variance contrast and the frame difference contrast are selected for generating the 2D spatiotemporal entropy. The extracted object is first represented by a group of(4 x 4) blocks coarsely,then the intra-block edge extraction on edge blocks and the joint spatiotemporal similarity test among neighboring blocks are further performed for determining the meaningful real objects. This proposed hierarchicalsegmentationalgorithm may be very useful for MPEG-4 applications.A novel fast algorithm is also introduced for reducing the search burden. Moreover,this unsupervised algorithm also makes the automatic image and videosegmentation possible.
暂无评论