Non-blind post-processing algorithm for remote sensing image compression

Abstract

High-efficiency compression of remote sensing images (RSIs) is very necessary after images are acquired because the on-orbit transmission bandwidth and memory capacity are limited. Wavelet-based compression methods have been widely used in on-orbit image compressors for optical cameras. However, wavelet transforms have low sparse representation capability for edges (i.e., high-frequency information) in RSIs. A lot of wavelet coefficients of edges have high magnitude because spatial redundancies between these coefficients still exist, which is not​ suitable for the subsequent compression. In this paper, we propose a non-blind post-processing approach in the wavelet domain. The non-blind post-processing uses a high-frequency detection algorithm to establish a high-frequency map, which is used to directly guide the allocation of post-transform resources (e.g., multi-basis dictionary post transform and the rate–distortion estimation). Post-transform resources can be allocated to high-frequency areas but not to low-frequency areas because the smooth areas need not be performed by the post-processing, while detailed areas need more post-processing resources. The best transform estimators are only performed to determine the best transform at the high-frequency areas, while need not at low-frequency areas. The proposed method can improve the post-processing efficiency and compression performance because the post-transform exploits the redundancies among wavelet coefficients and removes large-amplitude coefficients of high-frequency areas in the wavelet domain. The proposed method is confirmed and experimental results demonstrate that the proposed method obtains a high calculation efficiency and high compression performance compared with the blind post-processing. The proposed method is suitable for the compression of RSIs and other images.