Adaptive Predictive Image Coding Using Variable Block Shapes Controlled by Edge Orientation

This paper proposes adaptive predictive image coding with variable block shapes based on edge orientation, as a highly efficient coding method for low bit rates. the method aims at the improvement of reconstructed image quality in edge areas. In addition to the conventional method, where the quantizer is switched adaptively, according to the standard deviation of the prediction error signal calculated for each block 8 × 8 pixel, a predictor matched to the edge structure is used. Whenever necessary, the block is segmented into two, according to the orientation of the edge, and vector quantization is applied to the prediction error signal. When the block is segmented, the maximum number of bits assigned to each pixel is doubled if the number of representative vectors in quantization is kept the same. By this scheme, saturation of the signal-to-noise (SN) ratio for the high bit-rate condition, which is observed in the conventional system, can be avoided. In the proposed system, the forementioned control is executed by table look-up based on the standard deviation of the prediction error signal. the table is independent of the image and can be constructed beforehand with the advantage of the conventional system being inherited. the proposed method exhibits overall satisfactory coding performances over A-VQ-PE and A-DCT-VQ, which have high coding efficiency. This verifies the usefulness of the proposed method.

[1]  Robert M. Gray,et al.  An Algorithm for Vector Quantizer Design , 1980, IEEE Trans. Commun..