Astronomical Image Coding Based on Graph Fourier Transform

Lingshan Li, Yan Zhao, Shigang Wang

Abstract: In recent years, astronomical image coding has attracted increasing attention. The existing image compression algorithms are usually developed for ordinary images, which ignore the image characteristics and storage purpose of astronomical image itself, resulting in low compression efficiency. Aiming at the existing problems, we proposed an astronomical image compression algorithm based graph Fourier transform (GFT), which is mainly devoted to the high performance compression of the astronomical image with a deep space background taken by the ground astronomical telescope. The algorithm not only improves the compression ratio of the image, but also better preserves the information of the targets, so as to realize the storage of a large number of high-resolution astronomical maps in the limited storage space. Firstly, the GTF basis dictionary is constructed according to the result of the classification of astronomical image blocks by Weisfeiler-Lehman (W-L) subtree kernel. Then, during image block coding, the transform basis of the same kind of images is selected for GFT according to the calculation of image similarity, and different quantization matrices are adopted for quantization operation. Finally, the quantized transformation coefficients and the dictionary indexes are encoded by run length encoding and Huffman coding. By comparing with the image coding standard, it is verified that the proposed algorithm has higher peak signal-noise ratio and structural similarity index at low pixel depth than the existing image and video coding standards, and has better compression performance.