Abstract: Due to the non-stationary property of interferential multispectral image data, a novel compression algorithm for interferential multispectral images with proposed Empirical Data Decomposition (EDD) is presented. EDD can make a multi-resolution analysis of the non-stationary interferential data. With its local characteristic and variation tendency, the non-stationary interferential data are decomposed by EDD into two parts: the sum of local region data and the difference data. In this paper, EDD is first utilized for interferential multispectral image data de-correlation, and a corresponding 2-D decomposition structure is presented as well. The decomposition coefficients are finally coded with the modified EBCOT. Experimental results show that, compared with the JPEG2000 standard, the proposed algorithm decreases the average output ratio by about 0.15 bit／pixel for lossless compression, and improves the reconstructed images by 1.1~2.5dB. The algorithm also reduces the Relative spectral Quadratic Error(RQE) and protects the spectral information efficiently.

Key words: image compression, interferential multispectral image, empirical data decomposition