基于信息量失真测度的VQ及在高光谱图像无损压缩中的应用

A VQ Based on Information Distortion Measure and Its Application to Lossless Compression of Hyperspectral Image

随着成像光谱技术的迅速发展 ,如何高效无失真的压缩海量高光谱数据引起人们越来越多的关注。由于相似的地表区域具有相似的光谱曲线 ,矢量量化是对高光谱图像进行压缩的理想算法。提出一种基于信息量失真测度的矢量量化编码方法 ,并用于高光谱图像无损压缩。与常用的矢量量化失真测度———欧几里德平方误差测度相比 ,该算法在不增加运算复杂度的情况下 ,矢量量化后的误差图像的熵值能够降低 0 0 5bpp左右。

The volume of image data generated by airborne and spaceborne remote sensing mission have been increased dramatically. The efficient lossless compression is urgent. A hyperspectral image comprises a number of bands, each of which represents the intensity of return from an imaged scene received by a sensor at a particular wavelength. Since the reflectance of the earth's surface and atmospheric absorption are wavelength dependent, the brightness vector formed in the spectral domain for each pixel will have a similar form. The relationship between type of ground and spectral response means that a hyperspectral image can be considered as a group of brightness vectors. Therefore VQ(vector quantization) an the ideal candidate for compression. If VQ is used to compress image losslessly, both the codevector index and the quantization error image should be sent to channel. The amount of codevector index is invariable, consequently, it is important to reduce the error image's average information amount, i.e. entropy, if we want to improve the coding efficiency. In this paper, a new VQ lossless compression method based on an information distortion measure is proposed. Using this new measure to match codevector, i.e. quantize vector, the coding efficiency can be improved without increasing complexity. Experimental results show that the entropy of the error image using VQ based on information distortion measure is about 0.05bpp (bits per pixel) lower than that on Euclidean square error measure.