Research On Stereoscopic Images Quality Assessment Methods

Stereoscopic image can give human stereoscopic perception and real scenes feeling just like you are on the scene,and gradually become the mainstream direction of research.The stereoscopic image market demand for application and development is more and more urgent.Stereoscopic image quality assessment method is one of the important property indexes to systems,so the research of stereoscopic image quality assessment method consistent with human eye's subjective feelings is meaningful.The paper proposed several stereoscopic image quality assessment methods which is on the basis of color information,besides,it further added edge information,edge structure similarity information and texture information.The main research contents are as follows.Firstly,according to color information plays an important role in stereoscopic image quality assessment,the paper proposed color stereoscopic image quality evaluation model based on HSV color space.The method extracts the color features of H,S,V,which is based on HSV color space,and use two kinds of distance metrics to measure the differences of the features between reference images and distortion images,analysis the importance of color feature to DMOS.Experimental results show that the Pearson Linear Correlation Coefficient(PLCC)and Spearman Rank Order Correlation Coefficient(SROCC)between the model and DMOS achieved 0.873 and 0.834.The results verify the color information has a very important influence on the stereoscopic image quality.Secondly,color stereoscopic image quality assessment method based on edge information,which is on the basis of color information and it further added edge information in different directions.Extracting edge information in different directions by using the Kirsch operator,and using cosine distance to measure the differences of the features between reference images and distortion image as edge quality indexes.And then,using color quality indexes and edge quality indexes construct stereoscopic image quality evaluation model.Experimental results show that PLCC and SROCC between the model and DMOS achieved 0.876 and 0.842.The results prove that the assessment model has a good performance on stereoscopic image.Thirdly,the quality evaluation method is on the basis of extracting color and edge information and it combine the structure similarity method(SSIM).Calculating structural similarity value of edge features between reference images and distortion image.The structural similarity results as edge quality indexes.And then,using color quality indexes and edge structural similarity results construct stereoscopic image quality evaluation model.Experimental results show that PLCC and SROCC between the model and DMOS achieved 0.889 and 0.859.The results show that the assessment model has a good performance on stereoscopic image.This method is superior to previously proposed method based on edge information.Finally,color stereoscopic image quality assessment method based on texture information,which is on the basis of color information and it further added texture information.Extracting txture information by using 5 ? 5 Laws texture templates,using cosine distance to measure the differences of the features between reference images and distortion image as texture quality indexes.And then,using color quality indexes and texture quality indexes construct stereoscopic image quality evaluation model.Experimental data show that PLCC and SROCC between the model and DMOS achieved 0.923 and 0.917.This method is superior to previously proposed method.For all evaluation method that put forward was tested on LIVE 3D Image Quality Database,which included both symmetric and asymmetric distorted 3D images published by university of Texas.This paper analyzes the scatter diagram and correlation coefficient which between image quality predictive value and DMOS.And then compare the results of the method proposed in this paper with the results of method that proposed by other researchers,analyze and compare the pros and cons of each performance.