Performance Improvement of the Adaptive KL-Transform Image Coding Based on Variable-Shape Blocks

Abstract

We previously reported on development of an adaptive transform coding scheme which partitioned an image into variable-shape blocks so that their boundaries could run parallel to the image's principal contours. That scheme also employed an adaptive-smoothing filter which smoothly connected a DC component in each block. Therefore the scheme could diminish interference such as mosquito-noise and blocking-effects. In order to further improve coding performance, this paper proposes two methods. One is related to segmentation, and magnifies movable areas of vertices of blocks to fit more accurately their boundaries to the image's contours. The other one introduces directional autocorrelation models for the purpose of adaptively deriving eigenvectors in each block. Simulation results demonstrate that the proposed scheme using these two methods provides 1-2 dB and 2-3 dB additional improvement in the signal-to-noise ratio over the previous scheme's and JPEG's levels respectively, and reproduces better images than either of these two.