Adaptive KL-Transform Coding Color Images Based on Variable-Shape-Blocks.

This paper proposes a new transform coding scheme which partitions a color image composed of luminance signals and two color difference signals into variable-shape-blocks. This partition reduces interference such as mosquito noise and blocking effects. The shapes of the blocks for the above three signals accord with one another in order to reduce the amount of computation required for image segmentation and derivation of basis vectors in a KL-transform, as well as the information load related to the block shapes. This scheme attains segmentation and bit allocation for color images by minimizing the power of residual signals which are obtained through the application of a new adaptive smoothing filter, and the final mean square error in the RGB signal space. Simulation results demonstrate that this scheme improves SNR to about 2 dB more than what is attained by a JPEG Baseline System, and reproduces better images with less interference.