Contrast-Aware DCT for Image Enhancement with JPEG Compatible Coding
Kohei Hayashi, Soichiro Honda, Hirokazu Kamei, Yoshihiro Maeda, Norishige Fukushima · 2024
JPEG is a de facto standard image coding format that is still used today. Although JPEG has a lower coding efficiency than new coding methods such as WebP, HIEF, and JPEG-XL, it has a high rate of adoption; thus, there have been several proposals for improved JPEG coding that maintain backward compatibility, such as Guetzli, MozJpeg, and Jpegli. These improved JPEG codecs optimize the quantization table but do not change the discrete cosine transform (DCT). Recent advances in signal processing have led to various extended DCTs. The use of redundant DCTs, which a type of extend DCTs, has made it possible to achieve a wide range of processing. In this paper, we propose a method for extending the redundant DCT to a contrast-aware process and for achieving DCT suitable for image enhancement while maintaining backward compatibility for JPEG. A naïve implementation requires 64 times DCT per block, but this paper proposes a method for calculating it quickly. The experimental results showed that the coding efficiency improved by 2.57 %, and significantly suppressed coding noise when enhancing.