|
Vector quantization (VQ) is an efficient technique for image compression. An new fast VQ scheme, called PCA-TSVQ, based on principal component analysis (PCA) is proposed. The proposed algorithm uses the first principal components of training vectors to construct a binary tree classifier. A simple thresholding method on the first principal components is used to find the split key on each nonterminal node. The centroid of training vectors belonging to the same leaf of the binary tree forms a codevector of the codebook. A modified scheme based on PCA-TSVQ is also developed. The modified scheme generates an adaptive codebook for a given image and encodes this image simultaneously as soon as the codebook is generated. Experimental results show that the new algorithm requirs about 2.5 secodes on a Pentium-150 PC to encode a 512 × 512 gray level image to archieve compression ratio 0.563bpp with satisfactory image fidelity.
|