Image Quality Assessment Based On Visual Perception

With the development of visual communications and information technologies,image,which is a simple and convenient carrier of visual information,has becoming increasingly important to humans' life.However,during the process of acquisition,transmission and storage of images,image distortions commonly exist and reduce the visual quality.Such degradation of visual quality has negative effect on the subsequent application of images.Therefore,image quality assessment(IQA),which aims to pre-dict the quality of images,has become a vital yet challenging task for digital image applications.Humans are the ultimate users of images.Thus,how to measure the image quality from the view of humans is always the key content of IQA.According to the above analysis and previous studies of IQA,this dissertation explores the reasonable visual perception models that are consistent with the characteristics of human visual system(HVS),and studies IQA methods based on visual perception to further improve the consistency between objective IQA method and subjective image quality evaluation.The main works of this dissertation are introduced as follows.1.A structural variation detection method based on binary logic is proposed.Based on the threshold behavior and just noticeable difference,this method transforms struc-tural features into the existences of corresponding features.Then,by comparing the existences of structural features between original and distorted images,the structural variations are detected.The binary logic based method experimentally outperforms the other traditional structural variation detection methods.In addition,based on the struc-tural variation detection method,this dissertation proposes a simple full-reference IQA(FR-IQA)method,which predicts the image quality by combing the structural distortion and the luminance distortion.The structural distortion is calculated only on the areas of images that are detected as having structural variations.According to the experiments,the proposed FR-IQA method achieves satisfactory performance.2.A perceptual model of structural variation classification is proposed.This model classifies structural variations/distortions into four categories:slight deformations,ad-ditive impairments,detail losses,and confusing contents.Each category has a unique appearance.This perceptual model was verified by a subjective experiment.The pro-posed perceptual model is achieved by using fuzzy-logic classifiers and majority voting strategy to analyze a set of structural features extracted from images.Based on the struc-tural variation classification model,a novel FR-IQA method is proposed.This proposed FR-IQA method evaluates the image quality by combining two aspects:the distribution of different structural variations and the degree of structural differences.The experi-mental results compared with other state-of-the-art IQA methods on seven public IQA databases indicate that the proposed method yields promising assessments.3.A novel no-reference IQA(NR-IQA)method for JPEG images based on block-iness and the luminance change is proposed.It has been reported that the vast majority of NR-IQA methods for JPEG images predict the image quality only by measuring the blockiness.However,on the basis of previous methods,the proposed method includes luminance change,which is a new index,as a supplement to blockiness.Based on the attention mechanism of visual perception,the proposed method obtains the more signif-icant regions of JPEG images.The proposed method obtains the final quality of JPEG images by weighted combining the blockiness and luminance changes extracted from the detected regions.The experiments compared with other state-of-the-art NR-IQA methods on the JPEG image datasets demonstrate that the proposed method has lower time complexity and better performance.4.An efficient NR image sharpness assessment method is proposed by combining the maximum gradient and the variability of gradients of blurry images.According to characteristics of visual perception,the subjective perceived sharpness is usually judged from the sharpest region of an image.However,the proposed method evaluates the im-age quality not only from the sharpest area of a blurry image,but also from the overall content variations in a blurry image.In the proposed method,the maximum gradients are used to reflect the sharpest information and the variability of gradients are used to de-scribe the overall content variations.The final quality is obtained by weighted combing the two components mentioned above.This dissertation compares the proposed method with other state-of-the-art NR image sharpness assessment methods on commonly used blurry image datasets.The experiment reveal that the propsoed method has extremely low time complexity and relatively high consistence with the subjective evluations.