On a new adaptive quantization scheme in subband-based video coding

This paper presents a new adaptive quantization scheme for subband video coding using wavelets. We aim at designing a quantization that is both signal adaptive and scene adaptive. By scene adaptive, we mean that the quantization is able to shift its bit allocation to perceptually more important coefficients so that the preservation of scene structure can be achieved. This new quantization scheme is based on the principle of adaptive clustering with spatial neighborhood constraints characterized by the Gibbs random field. With such spatial constraints, it is able to reach beyond the limit of the traditional simple statistical models that all lack of scene adaptivity. In addition, the dynamically enforced spatial constraints are able to overcome the artificial block division which is often the major source of distortion when coding at low bit rate. We have successfully applied this scheme to videoconferencing and obtained very favorable results. We believe that this new adaptive quantization will have an impact on many other applications because it is able to perform simultaneously signal adaptive and scene adaptive quantization.