Progressive compression based on multiscale edge model

Abstract

Wavelet-based representations of images are superior to traditional block-based methods due to its unique joint space-frequency characteristics, particularly for low bit rate coding. However, in wavelet-based image coding, the quantization errors I the high frequency subbands where the sharp structures are present result in various distortions such as contouring effect, graininess, and blotches in smooth regions, the ringing effect and the blurring effect near sharp edges. In order to overcome the artifacts, a numerous postprocessing techniques were proposed. However none of them could completely conquer the artifacts inherently caused by wavelet-based coding. Edges play an important role in visual perception. In this paper, we propose an image compression and reconstruction scheme based on the multiscale edge model with a low bitrate adaptive to various applications. The multiscale edge model decomposes and represents the image with its multi-scale primal sketch and the background. Multi-scale primal sketch is the union of the pulse edges and the ramp edges extracted at consecutive scales. Background is the small image survived after a hierarchical scale transform of edge-removed image. The multiscale edges are then encoded using the hybrid run-length and the generalized finite automata methods to preserve the significant edge structures while achieving desired bit budgets. Experiment shows that the proposed scheme can achieve good compression ratios while preserving comparable signal-to-noise ratios.