Multispectral Image Compression Using Interband Correlation And Wavelet Transform

研究了六波段Landsat专题成像仪图像的两种高效无损压缩方法。测试使用了小松机场周边地区、石川县的田野地区和山区的图像。带内预测与带间可逆小波变换。首先，去除各波段像素间的空间相关性，进行二维预测，使图像熵平均降低25%左右。其次，为了消除带间相关性，对同一空间位置的两个最短波长波段及其相邻波段的样本进行RW变换，传输高频（HF）样本。低频（LF）样本再次与下一长波波段的样本进行两点RW变换，并传输高频（HF）样本，以此类推。这样，熵就减少了0.11 bpp2。带内RWT和带间可逆KL变换（RKLT）首先，将每个带图像的两层RW变换成7个子带。此外，对最低子带进行二维预测。图像的熵平均降低约28%。其次，对6幅波段图像的同频子带样本进行六点KL变换；然后，将KLT矩阵的行相互替换，使得在每一行中，对角元素是最大值；由此导出RKLT矩阵。之后，将第二层子带的特征图像根据方差分为8类，熵减少约0.2bpp。

Two methods of efficient lossless image compression are investigated for six band Landsat Thematic Mapper imagery. The images of the region surrounding Komatsu airport, a field region and a mountaneous region in Ishikawa Prefecture are used for testing.1. Intraband prediction and interband reversible wavelet transform(RWT).First, to remove the spatial correlation between pixels within each band, 2D prediction is performed, by which the entropy of the images decreases about 25% in average. Secondly, to remove interband correlation, two samples of the shortest wave length band and its adjacent band in the same spatial position are RW transformed and the high frequency (HF) sample is transmitted. The low frequency (LF) sample is again 2-point RW transformed with the sample of the next long wave length band, and the HF sample is transmitted and so on. By this, the entropy decreases by 0.11 bpp.2. Intraband RWT and interband reversible KL transform(RKLT)First, each band-image is two layer RW transformed into seven subbands. In addition, for the lowest subband, 2D prediction is performed. The entropy of images decreases about 28% in average. Secondly, the same frequency subband samples of six band images are six-point KL transformed. Then, rows of KLT matrix are replaced each other such that, in each row, the diagonal element is the maximum; from it, RKLT matrix is derived. After that, the eigen images of the second layer subbands are classified by their variances into 8 classes, by which the entropy decreases about 0.2bpp.