Perceptually Based Downscaling of Images

摘要

We propose a perceptually based method for downscaling imagesrnthat provides a better apparent depiction of the input image. Wernformulate image downscaling as an optimization problem where therndifference between the input and output images is measured usingrna widely adopted perceptual image quality metric. The downscaledrnimages retain perceptually important features and details, resultingrnin an accurate and spatio-temporally consistent representation ofrnthe high resolution input. We derive the solution of the optimizationrnproblem in closed-form, which leads to a simple, efficient andrnparallelizable implementation with sums and convolutions. The algorithmrnhas running times similar to linear filtering and is ordersrnof magnitude faster than the state-of-the-art for image downscaling.rnWe validate the effectiveness of the technique with extensiverntests on many images, video, and by performing a user study, whichrnindicates a clear preference for the results of the new algorithm.