The Research Of No Reference Image Quality Assessment For Multiply Distorted Color Images

In modern society,digital images,as an important medium for carrying information,are increasingly becoming indispensable components for communication in life,work,and study along with text and speech.Moreover,due to the rapid development of multimedia technology,digital images are becoming more and more widely used in life,scientific research,industry,and other fields.However,some distortions are often introduced in the process of image acquisition,compression,transmission,and storage making the image quality unable to meet the expected requirements,thereby affects the subsequent processing or use,such as image enhancement,face recognition,and super-resolution.In recent decades,objective image quality assessment(IQA)algorithms aimed at reflecting the subjective quality perceived by humans have been greatly developed,and many IQA algorithms have been proposed.Although these algorithms are already able to efficiently evaluate the quality of single distorted images,most of them perform poorly when faced with multiply distorted images.In this paper,the research of no reference(NR)quality evaluation for multiply distorted images is studied,and two IQA algorithms for multiply distorted color images are proposed based on the recent brain research and the visual perception characteristics of the human visual system(HVS).The latest research on human brain theory shows that the brain has an internal generative mechanism(IGM)when processing visual information.This mechanism indicates that the brain will actively predict the primary information of the received signals and compress the uncertainty information simultaneously.According to IGM theory,a new no reference image quality assessment algorithm by information decomposition is proposed in this paper.The image is decomposed into the primary part containing primary information and the residual part containing uncertaintyinformation.Then,the primary features,the residual features,and the global features are encoded using the local binary pattern(LBP)to describe the distorted information of different types in the image.Another study shows that the human visual system has different sensitivities to different frequency components when perceiving image information,and the influences of different distortions on the low-frequency and high-frequency components are quite different.For example,noise mainly affects high-frequency components,while blur mainly affects low-frequency components.Therefore,another no reference image quality approach by distortion separation is proposed in this paper to improve and extend the aforementioned algorithm.The image is decomposed into the orderly part composed of low-frequency components and the disorderly part composed of high-frequency components.The orderly features,the disorderly features,and the global features are then encoded using the local derivative pattern(LDP)to describe the distortion information of different frequencies.The experimental results show that the two IQA algorithms proposed in this paper achieve significant progress in the efficiency,stability,and time complexity in evaluating the quality of multiply distorted images compared with other algorithms of the same type.