Removing JPEG blocking artifacts using machine learning

JPEG is a commonly used image compression method. While it normally yields very good compression ratios, it also introduces blocking artifacts and quantization noise. In this paper, we present a method to remove noise and blocking effects from JPEG-compressed images. We use machine learning techniques to predict DCT coefficients and pixel values in a compressed image. Results show a decrease in mean square error between our predicted images and the original uncompressed images when compared to the compressed images, as well as a clear reduction of blocking artifacts.