For reversible compression of multispectral images, we use a technique consisting of 2-D intraband (spatial) two layers reversible WLT for each band image, followed by interband M-points reversible KLT, where M is the number of bands (in this paper, M=6) and each subband uses its own KLT matrix. In the RKLT, it is often observed that the RKLT coefficients significantly differ from real KLT coefficients, resulting degradation of compression efficiency. To overcome this problem, we propose to permute the rows of a KLT matrix so that each diagonal element has the greatest amplitude in its own row. By this, the degradation becomes negligible small. We named it quasi-diagonalized RKLT (DRKLT). Experimental results on five sets of multispectral images show entropy reduction of 1 27bit/pel in average for RWLT and of 0.32bit/pel in average for DRKLT.
展开▼