A New Image Downscaling Algorithm based on a Circular Area Pixel Model
Su Hyon Kim · 2022
Image downscaling is a fundamental task for most image and video applications, and many research works have been proposed. Although most of them try to get a scaled image of higher quality and to reduce the processing time further, to develop more and more efficient algorithm is still challenging. In this paper, we propose a new method for image scaling based on a circular area pixel model rather than a rectangular area pixel model. In case of upscaling, both an original and a target pixel are treated as circular regions whereas the target pixel is treated as an elliptical region only for downscaling. A pixel's weight represents a spatial contribution of the original pixel to a target pixel by the area of their overlapped region. Compared with existing algorithms by experiments, the proposed algorithm produces downscaled images of better quality than those obtained by the others including Bicubic and Lanczos. The proposed filter kernel can be adopted as a spatial kernel for the existing edge-preserving image downscalers to improve their performance further.