Tree-Structured Vector Quantization with Flexible Multipath Searching Method Based on Side Match Prediction

Although multipath tree-structured vector quantization (MP-TSVQ) has a higher probability of determining the closest codeword for each input vector than singlepath TSVQ(SP-TSVQ), it lacks flexibility in some cases because of the fixed number of search paths. To overcome this drawback, we propose the side-match prediction TSVQ (SMP-TSVQ) to make the number of search paths adjustable. SMP-TSVQ uses the correlation of adjacent blocks, and the smoothness of the upper block and the left block relative to the current block to determine the number of search paths. For seed blocks that play an important role, 8-path is always used to guarantee the quality of the image. For residual blocks, there are three cases to be considered, i.e., 1) when the upper block and left block are both smooth, 2-path is used; 2) when either the upper block or the left block is smooth, 4-path can be used; 3) when neither block is smooth, 8-path must be used to ensure the quality of the image. Therefore, compared to the previous multipath search algorithm, SMP-TSVQ requires flexible search paths to reduce computational complexity and encoding time. Experimental results have proved the expected merits of the proposed scheme. Moreover, when the codebook sizes are 128, 256, 512 and 1024, the encoding time of SMP-TSVQ is only 34.9%, 20.3%, 13.9%, 6.7% of that of FSVQ, respectively.