Image Quality Assessment On 2D And 3D Image

With the development of computer technology and social information,digital image has been widely used in all aspects of human society.With the development of computer technology and social information,digital image has been widely used in all aspects of human society.As one of the key issues in the development of image applications,image quality assessment has become a hot research topic in the field of image processing.The purpose of the study is to construct an image quality assessment model by extracting the image feature so that the results of image quality assessment obtained by the model are consistent with the subjective perceptions of human visual system.At present,many effective 2D image quality assessment metrics and 3D image quality assessment metrics have been presented,but the correlation between the assessment metrics and human visual perception can still be improved.This paper is devoted to the study of 2D and 3D image quality assessment algorithms.The specific works are as follows:Firstly,this thesis presents a saliency-aware 2D image quality assessment algorithm.Image saliency detection technique simulates the principle of visual saliency to automatically recognize the salient region in the image.Since human visual system is more sensitive to the salient region in the image,we propose an algorithm to integrate image saliency detection into 2D image quality assessment.In order to study the issues of image saliency detection and image quality assessment at the same time,we create the image dataset TID2013S.The experimental results on TID2013S show that our saliency-aware approach improves the traditional 2D assessment metrics.Secondly,this thesis proposes a pairwise-based learning-to-rank 2D image quality assessment algorithm.Existing 2D image quality assessment metrics perform differently in different image datasets.To solve this problem,we present to combine the advantage of each assessment metric by using the learning-to-rank method based on the pairwise-based preference feature.The experimental results on some 2D image quality assessment datasets show that our assessment metric outperform 12 2D image quality assessment metrics.Thirdly,this thesis presents a multi-cues 3D image quality assessment algorithm.In recent years,3D image and video applications have developed rapidly,but there are only few works on 3D image quality assessment.Current 3D display technology utilizes the disparity between the left and right eyes to reproduce the stereo vision in the brain.So 3D images are associated with three clues:left view,right view and disparity map.We propose to combine these three cues to design the 3D image feature and use the pairwise-based learning-to-rank algorithm to assess 3D image quality.The experimental results on 3D image dataset show that this approach obtain good performance than 12 2D-based 3D image quality assessment metrics and 3 specially designed 3D image quality assessment metrics.