Optimization Vector Quantization by Adaptive Associative-Memory-Based Codebook Learning in Combination with Huffman Coding

In the presented research on codebook optimization for vector quantization, an associative memory architecture is applied, which searches the most similar data among previously stored reference data. For realizing the learning function of new codebook data, a learning algorithm is implemented, which is based on this associative memory and which imitates the concept of the human short/long-term memory. The quality improvement of the codebook for vector quantization, created with the proposed learning algorithm, and the learning-parameter dependence of the improvement is evaluated with the Peak Signal Noise Ratio (PSNR), which is an index of the image quality. A quantitative PSNR improvement of 2.5 – 3.0 dB could be verified. Since the learning algorithm orders the codebook elements according to their usage frequency for the vector-quantization process, Huffman coding is additionally applied, and is verified to further improve the compression ratio from 12.8 to 14.1.