An Efficient Onboard Compression Method for Multispectral Images Using Distributed Post-Transform in the Wavelet Domain in Conjunction with a Fast Spectral Decorrelator
Jin Li,Zilong Liu,Shou-fu Tian
DOI: https://doi.org/10.1007/s10043-019-00492-9
IF: 1.2
2019-01-01
Optical Review
Abstract:A remote sensing multispectral image compressor must be of low-complexity, high-robustness, and high-performance because it is usually located on a satellite platform where resources, such as power, memory, and processing capacity, are limited. Multispectral images having multiple bands are mainly compressed using compression algorithms based on three dimensional (3D) transforms, such as the 3D discrete wavelet transform, which exhibits satisfactory compression performance. However, the principal compression algorithm used for multispectral images having relatively a few bands is to encode each band independently, without considering the spectral redundancy between bands, which results in low compression performance. In this paper, an efficient compression method for multispectral images having a few bands is proposed, which is based on a distributed, improved post-transform in conjunction with a low-complexity, fast spectral decorrelator. First, a fast spectral transform and an improved post-transform having only a fast principal component analysis basis are used to generate the spectral and spatial sparse representation. Second, a distributed, improved bit plane encoding is integrated into the post-transform to remove the remaining spectral and spatial redundancy. Experimental results show that the proposed approach improves compression performance for test data in different performance measures: peak signal-to-noise ratio, mean structural similarity index, and visual information fidelity. Compared with current state-of-the-art compression techniques, the proposed method exhibits a performance improvement of 0.3–1.7 dB PSNR.