Efficient compression algorithm using learning networks for remote sensing images

Abstract

Remote sensing image compression plays a vital role in the high-resolution imaging of an on-orbit optical camera. The post-transform-based compression method is of particular importance for remote sensing on-orbit images because it can remove remaining redundancies among high-amplitude coefficients in the wavelet transform, specifically in high-frequency areas. However, current post-transforms are inefficient because the post-transform has to access a large-scale wavelet domain. In this paper, we propose a low-dimensional visual representation convolution neural network (LVR-CNN) for efficient post-transform-based image compression. The LVR-CNN is used to transform the wavelet domain from a large-scale representation to a new wavelet version with a small-scale. We obtain the optimized small-scale wavelet representation by minimization between the original and reconstructed wavelet representations through LVR-CNN. The multi-basis dictionary post-transform is applied to the optimized wavelet representation to achieve high compression performance and calculation efficiency. We experimentally confirm the proposed method and results with test remote sensing images. The experimental results indicate that the LVR-CNN post-transform-based compression method yields high compression performance and low post-transform resource utilization. Compared with conventional methods, the proposed method can increase the peak-signal-noise-ratio (PSNR) by 1.2 dB∼2.7 dB. These merits indicate the proposed compression method is efficient for remote sensing images.