Perceptual Experiments Optimisation by Initial Database Reduction

摘要

Image quality plays important role in many image processing applications. For assessing perceptual image quality, there is need to quantify the visibility of differences between a distorted image and a reference image using a variety of known properties of the human visual system. Also to provide a convincing proof that a new method is better than the state-of-the-art the image quality assessment should be employed. Therefore image based projects are often accompanied by user studies, in which a group of observers rank or rate results of several algorithms. Unfortunately the problem posed by subjective experiments is their time-consuming and expensive nature. Huge size of input databases is crucial in that situation. This paper is intended to reduce the database size and made the subjective experiments less expensive and therefore more usable. To achieve it we employ a clustering technique and human visual system based objective metrics.