Region-based coding of images using a spline model

This paper describes a second-generation region-based coding scheme operating in a subband environment. Specifically, this is a compression technique based on approximating the lowest-frequency subband of an image with a 2-D spline model. Following a segmentation procedure, parametric polynomials are fitted to the resulting regions of the lowest-frequency subband. This subband is then represented by the spline coefficients, the region boundary description and the residual error. The residual error plus the higher-frequency subbands are encoded using conventional pixel-based quantization schemes while we describe an efficient encoding scheme for the 2-D spline representation. An entropy-constrained design procedure is developed for designing these individual encoders to minimize the overall mean-square distortion subject to a constraint on the output entropy. A comprehensive study of the operational rate-distortion performance of the resulting scheme is presented, which indicates the superiority of this approach to the more traditional pixel-based entropy-constrained schemes