Ringing artifact reduction for images resized in Block-DCT domain

Abstract

This thesis proposes an algorithm to reduce ringing artifacts caused by image resizing process in block-DCT domain. Ringing artifact is reduced by the proposed frequency domain filtering and spatial domain post-processing. While the previous ringing removal filters have to find a compromise between blurring and ringing artifacts, the proposed Algorithm introduces no further blurring and reduces the ringing artifacts. But new annoyance appears near the strong edges due to the amplified overshoot. The proposed method detects overshot regions and constructs a mask map indicating the degree of overshoot. Based on the mask map, the ringing artifacts can be reduced by combining the filtered image and unfiltered image in image domain. Experimental results demonstrate that the proposed algorithm achieves better subjective quality than the previous ringing reduction approach with little degradation of PSNR.