Abstract: For applications such as videophone or videoconferencing, we all agree that foreground object is more important than the background information. Nevertheless, in most conventional video codecs, such as H.261, due to the variation of the background lighting, noise or other unpredictable reasons, a large amount of available bandwidth is used to code the residue blocks at the background. Thus, it makes the coding inefficient when we consider the subjective visual quality of the foreground object. In this paper we proposed an object-based coding method. It tries to identify the foreground object and then put most of the resources to code it. We suggest a block-based DWT as a coding kernel because of its simple structure and outstanding performance. Experimental results show that we can compress a video sequence down to below 20 kb/s while having a reasonable visual quality.!12
展开▼