Fast Multi-level Image Vector Quantization

In this research work, a measure based on moment features is proposed to index the reproduction vectors of any VQ codebook. The rotation and reflection invariant feature of the proposed indexing measure was utilized to get further reduction in the vectors searching time. The quad tree coding scheme was adopted to partition the compressed images into blocks. The overall error caused by using the closest reproduction vector for representing each image block was utilized to decide whether the tested image block needs to be further partitioned or not. To improve the compression performance the coded blocks have been remapped such that their DC energy was set zero, and their AC energy is normalized (i.e., set equal to 1). The test results indicated that the proposed coding scheme leads to an encouraging compression results, and the required encoding time was greatly reduced without causing significant decrease in the image quality.