Image Quality Assessment Using Fractal Analysis

Ubiquity of digital cameras and smart-phones, especially the rapid development of socialnetwork and twitter, has resulted in an explosion of digital images in form of personal and in-ternet photo-collections—leading to a revolution of our daily life. Although such images forma well-established communication medium for sharing experiences or blogging events, it alsoposes challenging research questions. Since images are subject to distortions during acquisi-tion, compression, transmission, processing and reproduction, which will bring difficulties tothe subsequent process such as image and video processing, analysis and understanding. Thus,it is necessary to design or optimize an image and video processing method or system by mea-suring the degradation of the image and video, so that we can get the perfect visual quality withthe least cost.The aim of the image quality assessment is to find a computational model that canpredict the perceptual visual quality automatically.In this dissertation, objecting to essential problems of the image quality assessment andexploring the relation between human visual system and digital information, objective assess-ment methods are built to evaluate some distortions by uncertain factors in process of the imageand video, which are based on the image fractal analysis modeling for measuring visual qualityof distorted image and the ability to provide information of the image, The main contributionsof this dissertation can be summarized as follows:(1) According to fractal analysis of the image and mimicking mutil-channel characteristicof human visual system using Log-Gabor, fractal analysis is employed to extract geometricalcharacteristics of each subband of the image. The degradation of perceptual quality can be cap-tured by these fractal features. So a framework of reduced reference image quality assessmentbased on fractal analysis is proposed, thorough empirical studies are carried out upon the sev-en image databases against subjective mean opinion score and demonstrate that the proposedframeworkhasgoodconsistencywithsubjectiveperceptionvaluesandtheobjectiveassessmentresults can well reflect the visual quality of images.(2) According to fractal analysis of the phase congruency image, Scharr operator is usedto obtain the oriented subimage,the phase congruency map is getted of each oriented subim-age, fractal analysis is employed to extract geometrical characteristics of each phase congru-ency map. The degradation of perceptual quality can be captured by these fractal features.So a framework of reduced reference image quality assessment based on fractal analysis andphase congruency is proposed, thorough empirical studies are carried out upon the seven imagedatabases against subjective mean opinion score and demonstrate that the proposed framework has good consistency with subjective perception values and the objective assessment results canwell reflect the visual quality of images.(3) The Radon transform converts a pixel represented image into an equivalent, lowerdimensional and more geometrically informative Radon-pixel image. The projections of animage are first extracted by Radon transform. Then fractal dimensions are computed on eachprojection and concatenated as the image features that characterize the image structures fromthe view of the spatial distribution. Finally the image features are pooled as the quality scoreusing1distance. The proposed approach was evaluated on four public benchmark databases.Experimental results have demonstrated the excellent performance of the proposed approach.(4) Machine learning is used for the feature pooling process and demonstrated to be effec-tive. Thisistoaddressthelimitationsoftheexistingpoolingtechniques,likesimplesummation,averaging,Minkowski summation,etc., which tend to be ad hoc. The experiments show that theproposed method outperforms the existing relevant schemes. Extensive analysis and cross val-idation are performed with seven publicly available databases.