The Parallel Decorrelation Stretching with Multiple Decomposition Tactics for Remotely Sensed Imagery

This paper presents a parallel processing method of decorrelation stretching with multiple decomposition tactics for remotely sensed imagery. The method adopts different decomposition tactics for different steps in the whole procedure with band-based decomposition in the statistics of image bands, twin-band-based decomposition in the computation of the covariance matrix, and tile-based decomposition in the linear transformation. The whole procedure is parallelized. The parallel experiments of decorrelation stretching for two datasets, the airborne hyperspectral image OMIS and satellite image ASTER, are carried out on two multi-core computers respectively with Windows 7 and Linux operating systems. The results show that it can achieve whole-speedup up to eight on computers with cores ranging from 12 to 16 by correctly configuring the number of cores and disks. Meanwhile, the factors impacting the whole-speedup are analyzed, and usage suggestions for decorrelation stretching for remotely sensed imagery on the multi-core computer are proposed.