Panchromatic image compression based on improved post-transform for space optical remote sensors

Abstract

Panchromatic images of space optical remote sensors effectively demonstrate the shape, structure, and texture of landscape objects. Conventional compressors, such as JPEG2000 and Consultative Committee for Space Data Systems-Image Data Compression (CCSDS-IDC), widely use discrete wavelet transforms (DWTs) as sparse multiresolution representations of remote sensing panchromatic images (RSPIs). However, many large amplitude and high-frequency coefficients can be produced when RSPIs project onto a DWT basis, which are detrimental to the subsequent encoding process. Therefore, using only DWT to perform sparse representation is not an optimal solution for RSPIs. In this study, we propose a low-complexity compression approach based on a dual-loop, double-bases post-transform combined with bit-plane encoding (BPE) for RSPIs. First, a DWT is applied to RSPIs to perform the transform operations. Then, from the computed DWT coefficients a post-transform with a double base (discrete cosine transform and Hadamard transform) is performed. One of the double bases can be used for the high bit rate and the other can be used for the low bit rate. The best post-transform is selected based on a p-norms approach. The post-transformed coefficients are reorganized into tree structures to perform encoding using BPE. We use dual loop (basis loop and rate loop) to control the coding process. BPE results are fed back to select the basis used by post-transform. The best post-transform results are fed forward to allocate the bit rate for each coding segment. The experimental results of on-board RSPIs show that the proposed approach improves the PSNR by 0.6 dB to 1.1 dB compared to that of the CCSDS-IDC method with a small increase in complexity. The compression performance of the proposed method obtains similar results, but lower complexity to that of JPEG2000-based coders.