Tree-structured vector quantizers (TSVQ) and their variants have recently been proposed. All trees used are fixed M-ary tree structured, such that the training samples in each node must be artificially divided into a fixed number of clusters. This paper proposes a variable-branch tree-structured vector quantizer (VBTSVQ) based on a genetic algorithm, which searches for the number of child nodes of each splitting node for optimal coding in VBTSVQ. Moreover, one disadvantage of TSVQ is that the searched codeword usually differs from the full searched codeword. Briefly, the searched codeword in TSVQ sometimes is not the closest codeword to the input vector. This paper proposes the multiclassification encoding method to select many classified components to represent each cluster, and the codeword encoded in the VBTSVQ is usually the same as that of the full search. VBTSVQ outperforms other TSVQs in the experiments presented here.

Several tree-structured vector quantizers have been proposed previously. However, owing to the fact that all trees used are fixed M-ary tree-structured, the training samples contained in each node must be artificially divided into a fixed number of clusters. This paper presents a general-tree-structured vector quantizer (GTSVQ) based on a genetic clustering algorithm that can divide the training samples contained in each node into more natural clusters. Also, the Huffman tree decoder is used to achieve the optimal bit rate after the construction of the general-tree-structured encoder. Progressive coding can be accomplished by giving a series of distortion or rate thresholds. Moreover, a smooth side-match method is presented to enhance the performance of coding quality according to the smoothness of the gray levels between neighboring blocks. The combination of the Huffman tree decoder and the smooth side-match method is proposed herein. Furthermore, the Lena image can be coded by GTSVQ with 0.198 bpp and 34.3 dB in peak signal-to-noise ratio.

Although the side-match vector quantizer (SMVQ) reduces the bit rate, the image coding quality by SMVQ generally degenerates as the gray level transition across the boundaries of the neighboring blocks is increasing or decreasing. This study presents a smooth side-match method to select a state codebook according to the smoothness of the gray levels between neighboring blocks. This method achieves a higher PSNR and better visual perception than SMVQ does for the same bit rate. Moreover, to design codebooks, a genetic clustering algorithm that automatically finds the appropriate number of clusters is proposed. The proposed smooth side-match classified vector quantizer (SSM-CVQ) is thus a combination of three techniques: the classified vector quantization, the variable block size segmentation and the smooth side-match method. Experimental results indicate that SSM-CVQ has a higher PSNR and a lower bit rate than other methods. Furthermore, the Lena image can be coded by SSM-CVQ with 0.172 bpp and 32.49 dB in PSNR.

The full-search vector quantization suffers from spending much time searching the whole codebook sequentially. Several tree-structured vector quantizers had been proposed. But almost all trees used are binary trees and hence the training samples contained in each node are forced to be divided into two clusters artificially. We present a general-tree-structured vector quantizer that is based on a genetic clustering algorithm. This genetic clustering algorithm can divide the training samples contained in each node into more natural clusters. A distortion threshold is used to guarantee the quality of coding. Also, Huffman coding is used to achieve the optimal bit rate after the general-tree-structured coder was constructed. An experiment on speech coding was conducted. A comparison of the performance of this vector quantizer and the other two tree-structured vector quantizers is also given.