Research On Image Quality Assessment Based On Visual Perception Characteristics

Image is the main visual information carrier in information age.In the research of computer vision applications such as video conference,digital television,telemedicine,satellite communication,monitoring system,image quality is an important factor to measure the performance of the applications.So image quality assessment is one of the most important research topics in the field of image engineering.In this thesis,the image quality is mainly assessed from two aspects: image color and image definition.Now the research on no-reference assessment methods of image color cast and image definition is not perfect,this thesis has done some improvement and optimization for the existing methods' shortcoming.The experimental results show the effectiveness of the improved methods in this thesis.The main research work and achievements are as follows:1.Visual perception characteristics of image color and definition are extracted and simulatedFor the research on image color cast,at the based of the color constancy of vision system,the characteristic of visual brightness perception level is simulated by L component aggregation degree,and the characteristic of visual color perception migration is simulated by the parameters of ab two-dimensional histogram equivalent circle and the color cast factor in Lab color space.And the calculation methods of these parameters are studied and analyzed.For the research on image definition,the characteristics of luminance and chrominance adaptiveness,contrast and masking effect during perceiving image details are simulated,and the mathematical models of these characteristics are analyzed.In addition,the regional compensation effect is found by experiment,and its' calculation model is established.2.In the Lab color space,based on visual perception characteristics of image color,an improved assessment method of image color cast based on Lab color space is designedIn order to improve the accuracy of image color cast assessment,combining with the exiting assessment method of image color cast based on Lab color space,there are several optimization: In order to improve the time-consuming problem in the process of calculating characteristics,a new method of calculating the equivalent circle based on spatial aggregation strategy is presented;for the depth detection process of "initial normal image",in order to improve the precision of extracting range of NNO(near neutral objects)area(achromatic surface,whose essential color is white or gray),an adaptive NNO region extraction method is designed;by analyzing the original image's and its NNO region's parameters of ab two-dimensional histogram equivalent circle and the color cast factor,characteristic parameter Kch is introduced to improve the accuracy of assessment;for the reclassification process of “initial color cast image”,combining with the texture feature of the real color cast image and the intrinsic color cast image,texture richness is introduced on the basis of L component aggregation degree to optimize assessment condition to reduce the misclassification rate.3.Based on the visual perception characteristics of perceiving image details,an improved assessment method of image definition based on the retinal perception characteristic is designedIn the process of image definition calculation,the perception characteristics are not comprehensive in the existing assessment method of image definition based on retinal perception characteristic,and the assessment results do not match well with subjective perception.In this thesis,based on some physiological and psychological changes in the process of visual perception,an improved image definition assessment method based on retinal perception characteristic is proposed.A more comprehensive assessment model is established by integrating the original method with the masking effect,contrast sensitivity and regional compensation effect to enhance the consistency of the assessment results and subjective perception.4.The effectiveness of the two improved image quality assessment methods is verified by experimentsThe assessment methods of image color cast and image definition are verified in customized image database and accepted image databases,separately.In the experiment of image color cast assessment,the processes of initial classification,depth detection of "initial normal image" and reclassification of "initial color cast image" are analyzed.Then the experimental results are given and analyzed.Finally,making statistical analysis on the performance of the whole color cast assessment process,and comparing with several color assessment algorithms.In the experiment of image definition assessment,firstly,the thresholds of parameter c and ER of the regional compensation effect model are determined by experiment to simplify the calculation process of the assessment model;then the feasibility,validity and expansibility of the assessment method are analyzed on several accepted image databases,and the comparative data of the proposed algorithm and several definition assessment algorithms are given.The experimental results show that the improved assessment methods of image color cast and image definition have better accuracy and higher consistency with the subjective perception.