Research On Screen Image Quality Assessment Method And Construction Of Evaluation Platform

With the development of Internet technology,digital multimedia technology and big data cloud computing technology,screen images have received more and more attention.The practical applications of various screen images in digital entertainment,video surveillance and virtual reality have created practical value for all walks of life,which makes people put forward higher requirements for the quality of screen images.Due to the storage,transmission,display and other aspects of screen images inevitably lost information,resulting in low quality of screen images,not only affects the user experience,but also interferes with the subsequent screen image processing technology.In practical application,the researchers expect to use screen image calibration compensation method to improve the screen images’ visual quality,and only by making a reasonable evaluation of the screen images’ visual quality after correction and compensation,it can be judged whether the correction and compensation method is effective or not,so a reliable screen image quality evaluation method is needed urgently.At present,there are relatively few studies on the quality evaluation methods of screen images,and the performance of existing research methods is not ideal.Therefore,proposing and improving the quality evaluation method of screen image is helpful to quantify the distortion degree of screen images and further optimize the performance of existing screen image processing methods.According to the current research status of screen image quality evaluation,this thesis studies the reduced reference quality evaluation and no reference quality evaluation methods of screen images from two perspectives of screen content image distortion and screen image display distortion.The main work of the subject is as follows:(1)A reduced reference quality assessment method for screen content image distortion(RQASC)is proposed to realize the quality evaluation of screen content image distortion.Considering the huge difference between the screen content images and the natural scene images,the method innovatively proposes a multi-scale structure feature fusion method.First,the microstructures such as gradient magnitude and uncertain information of the screen content image,the macrostructure such as gradient magnitude and anisotropy,and color features are extracted to compare the distorted image and the reference image.Second,on the basis of multi-scale structural feature fusion,it proposes to fuse color features,remove unimportant details and retain important contour features to further improve the performance of the method.Finally,in the feature fusion stage,the gray histogram is used to the structural feature information is dimensionally reduced to reduce the degree of feature redundancy.In this thesis,the screen content image database SIQAD established by Nanyang University of Technology is used for experiments.The experimental results show that the proposed method has significantly improved the monotonicity,accuracy and consistency of evaluation compared with the fourteen most advanced screen image quality evaluation methods.(2)A no reference quality assessment method for screen image display distortion(NQASD)is proposed to realize the quality evaluation of screen image display distortion.Firstly,a new database of screen image display distortion is established.In order to improve the reliability of subjective scoring results,a new subjective quality evaluation method is proposed.Secondly,an objective quality evaluation method NQASD is proposed for screen image display distortion.The key factors of the screen image quality,find six characteristics that affect the visual quality of the screen image:contrast,complexity,brightness,saturation,color and naturalness.Finally,use the integrated support vector regression method to fuse the above visual features to get the screen image,which improves the robustness of the method.In this thesis,the newly established subjective evaluation database is used for experiments.The experimental results show that compared with the ten most advanced image quality evaluation methods,the proposed method has significantly improved the monotonicity,accuracy and consistency of evaluation.(3)Design and build a screen image quality evaluation platform.In order to use the above two methods in the real scene,this thesis uses Qt designer software to build a screen image quality evaluation platform.The basic functions of the platform include user registration,login,password retrieval and other functions.It also includes the call of the above two methods and the core functions of calculating quality scores and automatically exporting.It can quickly and accurately evaluate the quality of screen images,save manpower and material resources,and lay the foundation for the subsequent optimization of screen image processing algorithms.