LAUSR

Video web portals have to downscale an incredible number of videos and images to various resolutions each day. The latest image downscaling algorithms conserve edges and fine details but still suffer from noise amplification. This often makes undesirable artifacts especially with a large downscaling factor. Deep learning based methods show superior performance only for some predetermined integer factors. For a practical real-time image processing that meets the requirement of video portals, it is still challenging to scale image down as quickly as possible without emphasizing noise, preserving sharp edges and important details with an arbitrary non-integer and even a large factor. In this paper, we propose a new detail-preserving image downscaling algorithm based on inverse joint bilateral filtering using an area pixel model and moving average. The proposed method not only retains visually important details but also prevents unpleasant noise amplification and aliasing artifacts by smoothing filter coefficients with respect to the downscaling factor through two-step one-dimensional filtering. Experimental results show that the proposed algorithm is about 7.37% faster on average than the fastest existing detail-preserving image downscaler. GPU implementation of our algorithm downscales a 2K video to 128-pixel width without temporal artifacts at speed of 116 frames per second. Moreover, the PSNR and SSIM scores achieved by our method were respectively 35.9% and 16.5% higher on average than the highest values scored by the existing methods when downscaling images contaminated by 5% salt and pepper noise.