Local thresholding classified vector quantization with memory reduction

A new memory reduction method for classified vector quantization (CVQ) is presented. Symmetry reflection, rotation and inversion of edge subimages are used to join appropriate edge classes thus reducing the memory requirements for edge codebooks by 8(4) times for the classifier used in this paper. Besides the memory reduction, our method generates the more robust codebooks thus increasing the PSNR for images outside the training set. It also relieves codebook generation for high bit rate by reducing the number of images that should be inside the training set. The proposed method has been tested with a classifier that is based on the comparison of locally thresholded image vectors with a predefined set of binary edge templates.