INTRAFRAME IMAGE DECODING BASED ON A NONLINEAR VARIATIONAL APPROACH

Despite advances in the domain of source coding, little recent work ha s been devoted to the problem of joint coding and decoding. In particu lar, to our knowledge, the design of decoders has been little investig ated. In this article, we focus on decoding for intraframe images of v ideo data. In our approach, we propose a new method for decoding by no nlinear filtering. Here, we break with the usual approach of perfect r econstruction filters at the decoder, and instead pose the reconstruct ion model as a minimization problem. The solution can be viewed as an inverse problem with the optimization of the transform/quantization/de coding structure formulated using a variational approach. We introduce sufficient conditions on the design of the decoder involving a priori assumptions on the solution and knowledge of the coder (transformatio n and quantization). We develop an optimization method for the reconst ruction filters at the decoder to account for effects due to quantizat ion noise. Experiments using this nonlinear inverse dynamic filtering demonstrate peak signal-to-noise ratio gains over standard linear inve rse filtering as well as appreciable visual improvements. (C) 1998 Joh n Wiley & Sons, Inc.