Research On Vector Quantization And Its Application In Hyperspectral Image

Vector quantization (VQ) is an efficient technique for data compression. It hasbeen successfully used in speech coding and mage compression system, owing to itshigh compression rate, simple decoding process and minor distortion. It replaces theinput vector with the index of the most matched codeword and then transmits and storesthe index. While decoding, it only needs simple look up.With the aim at the application of vector quantization to image processing, thispaper expounds detailedly its elemental principle, relative conception, and presentdevelopment deeply explores its two key techniques: codebook design and fastcodeword searching. Then summarizes and analyzes the present typical algorithms andpresents modified algorithms.1In codebook design, with analysis of the traditional LBG algorithm, we know ithas some disadvantages, such as weak adaptive ability, sensitive to the initial codebookand high computational complexity. This paper presents an improved algorithm. It usesthe method of adjusting cell-centroid after the trained vector finds the matchingcodeword vector. This way can reduce the time consumption of codebook designing. Inaddition, it also introduces fast searching algorithm to remove unmatched codewords incodebook designing process. Simulation results show that the improved algorithm cannot only reduce the amount of computation, but also improve the codebookperformance.2On the basis of the existing codewords fast searching algorithm, this paperpresents two methods that consist of Hadamard transformation and codewordelimination inequalities that base on feature subvector. Because Hadamardtransformation is easy to get results and subvector can reduce codeword search regions.Compared with the original searching algorithm, these two improved algorithms cansignificantly reduce the codewords searching computation.3Hyperspectral images have the characteristics of large quantity of data and strongcorrelation between adjacent pixels in spectrum and space. Vector quantization isapplied to the compression algorithm of hyperspectral remote sensing image. Comparedwith LBG algorithm, the improved algorithm takes the image as whole to deal with. Ituses fast searching algorithm in the transition of different domains. The results show that the improved algorithm not only gets a reasonable compression radio but alsoincreases the quality of image and reduces the amount of computation.