Efficient Image Compression Through Differential Encoding of Super-Resolution Images.

Neural network image compresssion methods have experienced significant progress during recent years. However, one of the persistent challenges in these methods is the considerable time taken for encoding and decoding. This paper presents an innovative approach to address this issue by using Super-Resolution. The primary objective of our method is to reduce encoding and decoding times while ensuring the preservation of image quality, as measured by peak signal-to-noise ratio (PSNR) and multiscale structural similarity (MS-SSIM). Using the Kodak dataset, we demonstrate a significant reduction in encoding and decoding times and minimal changes in reconstruction data size while maintaining image quality.