Multispectral Image Compression Using Wavelet Transform

点击翻译按钮获取中文摘要

High efficiency lossy image comression is investigated for seven band Landsat Thematic Mapper imagery. The image of north part of Kanazawa city including Kahoku lake is used for testing.1.3-D DCT and scalar/vector quantizationRemoving Band 6 image, inter-and inter-band correlation between pixels in other 6 band images is removed by 3-D DCT.Band 6 image is solely compressed, because it differs greatly from other band images in spatial resolution. DCT coefficients are denoted F (i, j, k) where i : horizontal 8th order, j : vertical 8th order and k : band directional 6th order variables. As the coefficients with the same i, j and k has relatively small variance, they are united to form a block.In the DC block, the coefficients are DPCM-entropy coded. The other blocks are classified into 5 groups (G) according to the value of i+j+k. G1-G3 are scalar quantized and entropy coded, G4 is vector quantized with size of 1*4 and G5 is discarded.Bit-rate is abjusted by the ratio of the number of discarded small power vectors to the total number of vectors. In bit-rate greater than 1.2 bpp, PSNR greater than 32 dB is obtained.2.Wavelet transform and interband differential codingThe image of band 2 is two stage wavelet transformed and directionally vector quantized. Then differential pictures between the other band image and the reconstructed band 2 image are 3-D DPCM-scalar and vector quantized in the similar way to section l. As a result, 0.3 dB higher PSNR is obtained in the bit-rate of 1.2-1.6 bpp compared with those of 3-D DCT.3.2-CH perfect reconstruction IIR filter bankA filter bank system is developed such that all the analysis and Synthesis filters have linear phase and satisfy the power complementary condition approximately.