In view of the fact that spatial resolution and spectral resolution of remote sensing images taken by satellite becomes more and more greater, in some applications, it is needed to compress the multispectral images. An image registration method by phase correlation and affine transformation was proposed in order to improve the multispectral image compression quality. The proposed method effectively improved the correlation between the image spectrums. Aiming at the multispectral image compression problem, the Karhunen-Loève, KL transform method, which was used for eliminating correlation between spectrums in the image, and the low complexity two-dimensional wavelet encoding method were put forward. Compared with JPEG2000 independent compression method for each spectrum in an image, the Peak Signal to Noise Ratio, PSNR of decompression image by the proposed method improved 2.1 dB in average. Experimental results show that, under the same compression ratio, in this paper the proposed method can obtain better image quality than the JPEG2000 spectral image independentcompression method.