Automated image sharpening via supervised learning with human preferences

We propose an automatic method for image sharpening that maximizes the perceptual sharpness while preserving naturalness and original colors of a given image. We hypothesize a set of image properties to model the context for selection of sharpening parameters. We hypothesize and then verify that these properties contain the unknown feature (sub)space that could uniquely determine the best sharpening parameters. The (sub)space is learned through a training set of examples for which human preferences are obtained in psychophysical experiments. The human judgments are also used to learn the function that maps the (sub)space to the best sharpening parameter values. This function thus facilitates adaptive enhancement across an image since only the local image properties determine the value the function takes. Experimental results demonstrate the adaptive nature and superior performance of our approach over other algorithms. In addition, we present spatial approaches of respectively measuring the edge sharpness strength and the perceptual sharp- ness preferences, which do not require a reference image. The proposed approaches quantify the perceptual visual quality that reflects the responses of the human visual perception