Image Distortion Classification towards Quality Assessment Based on Tri-Training and Natural Scene Statistics

Abstract

An image distortion classification approach towards quality assessment is presented in this paper based on tri-training and natural scene statistics. At first, the semi-supervised learning of tri-training is employed to carry out the classification of different distortion images by the combination of labeled images with unlabeled images. Then the method of nature scene statistics is used to extract features of distortion images so as to lay a well foundation for effective classification. Through the synthetical integration of tri-training and nature scene statistics, a well effect of classification can be achieved. A series of experiment results show the performance advantages of the presented algorithm.