Font Size: a A A

No Reference Image Quality Assessment Algorithms Based On Natural Scene Statistics

Posted on:2017-01-07Degree:MasterType:Thesis
Country:ChinaCandidate:X M LuoFull Text:PDF
GTID:2308330485464013Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
People mainly through image information to obtain feedback from the outside world, has become the main way of people’s communicate. But the image in the processing of acquiring and transport due to the difference of the physical device and the difference image reconstruction algorithms, the difference will affect the quality of the image. Therefore, the primary issue is the effective evaluation for the quality of distortion image. Full reference image quality assessment,reduced reference image quality assessment and no-reference image quality assessment are the three main methods of distorted image quality evaluation. Unable to clear what type of distortion introduced in the image acquisition, transmission and storage process, it is in the practical application of the original image information is difficult to obtain, for this problem,the no-reference image quality assessment algorithm of image quality assessment without using any the original reference image information on the distorted image quality. In this paper, a natural image without reference information for this type of quality assessment conducted in-depth scientific research.BIQI algorithm is based on a two-step, modular non-reference image quality assessment algorithm, BIQI Firstly for the distorted image conduct wavelet transform to obtain sub-band factor, then the generalized Gaussian distribution model to obtain sub-band parametric factors. Parameters obtained in three dimensions, three directions to form a 18-dimensional vector, image feature vectors using the 18-dimensional vector representation. Then the image library distorted image (JPEG2000, JPEG, WN, G Blur, FF) are classified from SVM and feature vectors; last used feature vectors with the same kernel function and support vector machine to calculate the fitting Image Quality. However, this method for the evaluation of JPEG2000 compression distortion image and Fast Fading distortion image has a big difference form Subjective scoring DMOS; and compared with the full-reference image quality assessment algorithm PSNR it is not gifted.Based on the study after BIQI no-reference image quality assessment algorithm,on the type of the distortion of JPEG2000 image quality evaluation and the Fast Fading type the distorted image quality evaluation, conducting in-depth research and analysis, this paper based on natural scene statistics no-reference image quality assessment algorithm. The main work and achievements are as follows:(1) Based on the natural scene statistical of the wavelet domain no-reference image quality assessment algorithms. The BIQI of the evaluation of JPEG2000 compression distorted image quality evaluation algorithm of this module, using the statistical evaluation methods based on natural scene wavelet domain, the method for predicting the degree of deviation from the image quality by comparison between the distorted image with the reference image. Using this algorithm to obtain each sub-band of the two thresholds, and thus obtain four empirical probability, respectively located four quadrants prediction coefficients right. Finally, the weighted sum of the way to get the final image quality JPEG2000 compression artifacts.(2)Based on the natural scene statistical of the airspace no-reference image quality assessment algorithms. The BIQI algorithm of evaluation Fast Fading distorted image quality of this module, using the evaluation method of based on the airspace natural scene statistical. Image distortion by the local normalized luminance factor in the NSS features to measure, and then assessed by GGD and 18 parameters AGGD extracted due at different scales, distortion on the image is different, so the two on the scale, a total of 36 parameters can be extracted, the final quality Fading Fast distorted image by SVM and SVR fitting draw.(3) In this paper, the image database using LIVE image database, the image library by the JPEG2000 (1-227), JPEG (1-233), WN (1-174), G blur (1-174) and Fast Fading (1-174) class 5 and composed of in total 982 distorted image and 29 of the original undistorted reference image,experimental under Matlab2013a environment,results showed that:the improved algorithm in the evaluation of the entire database 982 LIVE image distortion, better than the original BIQI algorithm, especially in improve the JPEG2000 and Fast Fading type of distortion of the image quality evaluation, more similar to human scores DMOS subjective value.
Keywords/Search Tags:image quality assessment, natural scene statistics, wavelet domain, airspace, subjective score
PDF Full Text Request
Related items