Abstract.A simple and effective no reference blur image quality assessment algorithm based on wavelet high frequency singular value decomposition is proposed. As the different wavelet high frequency sub-bands in the same level are highly structural correlation, and the degree of correlation would be weaken as the degree of blur distortion strengthen. The proposed method decomposes the images by wavelet transform firstly. Then the singular value decomposition is used for different high frequency sub-bands to get their singular value vectors, which we used to represent their structural information. Thirdly the angles between different sub-bands singular value vectors are computed, which reflects their degree of correlation. Finally the sum of angles is used as the last objective assessment index. Compared to the traditional methods, the proposed algorithm is more efficient and practical as it does not need to train or create a reference image. Experimental results show its good effectiveness and performance on LIVE2, CSIQ and TID2013 databases. IntroductionNowadays digital images are widely used in various applications. But how to assess the quality of images objectively and effectively still remains a challenge [1]. Based on the availability of reference images, with which the distorted image is to be compared, objective image quality assessment (IQA) approaches can be classified into: full-reference (FR), no-reference (NR) and reduced-reference (RR) approaches. As the reference image is not available, no-reference approach is more widely used and has been the research hot spot in image quality assessment [2].The proposed method is a NR IQA method assumed to a blurry image. In this field, Rony Ferzli, etc[3] presented a perceptual-based no-reference objective image sharpness/blurriness metric by integrating the concept of just noticeable blur into a probability summation model. Xie Xiaofu, etc [4] introduced an no-reference structural sharpness(NRSS) method for quality evaluation of blurred images by constructing a reference image by a low-pass filter. Considering the human visual system, Yin Ying, etc [5] compute the blur value of image by producing a reference image through Gauss low-pass filter transformation and combining the singular value decomposition(SVD) and generalized regression neural network model techniques. These methods can be roughly reduced into two categories. The first category methods construct a reference image through some sort of filtering process and then evaluates the quality by comparing the difference between the before and after filtered image, such as in literature [4,5]. The second category methods firstly make some transformation to the image to get the feature information, and then predict image quality score by some kinds of machine learning models. The above two category methods needs either to construct a reference image or to train and learn with the opinion quality score by machine learning to predict the image quality, these will undoubtedly increase the computing cost and a...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.