A project has been published on GitHub , designed to remove compression artifacts from JPEG images. This is accomplished by restoring the accuracy of discrete cosine transform coefficients. However, within the limits allowed by the quantization table, it does not enable the algorithm to re-filter the image. The code is written in C with optimizations for SSE2, AVX2, and ARM Neon. Additionally, there is a WebAssembly version that allows for image processing in the browser (but is slower than the native version due to the lack of SIMD optimizations and multithreading).
According to the author, their algorithm works significantly faster than a similar project and produces clearer images. However, the author acknowledges that jpeg2png may perform better in complex cases. Another project, , published under the Google account — which is also supposed to remove JPEG artifacts, is claimed by the author to be "worthless" and "difficult to notice improvements from its work."
Source: opennet.ru
