Research On No-reference Blur Image Quality Assessment Algorithm

Objective blurred image quality assessment algorithm can be divided into full reference image quality assessment, reduced reference image quality assessment and no-reference image quality assessment. Compared with full reference and reduced reference algorithms, no-reference algorithm can predict the image quality straightly without any reference. It can be embedded in application system easily and has importance role in research study. This thesis proposed three kinds of no reference image quality assessment algorithms and their key points are written as follows:1. Proposed a no-reference blurred image quality assessment algorithm based on PSNR in DCT domain. The steps consists in implementing Gaussian blur for prediction image, and processing DCT transform for the prediction image and Gaussian re-blurred image, getting the DCT coefficient matrix; then extracting respectively the front100columns of DCT coefficient matrix of prediction image and Gaussian blurred image as image feature matrix, calculating PSNR between two image feature matrix in each column, obtaining feature vector DCTPSNR; Finally, using support vector machine regression (SVR) model to predict the quality of the images. The experimental results show that the algorithm performance is working well; it is much better than the traditional PSNR algorithm, and more useful than the traditional PSNR algorithm in the application.2. Proposed a no-reference blurred image quality assessment algorithm based on DCT coefficient histogram. The steps consists in doing DCT transform to get DCT coefficients histogram curve for prediction image; Then, using f (x) curves to express DCT coefficient curves in order to get the curvature formula; Finally setting the appropriate threshold to get image quality index according to the curvature formula. The experimental results show that the algorithm performance is superior on LIVE database and it is much better than the other no-reference image quality assessment algorithms.3. Proposed a no-reference blurred image quality assessment algorithm based on the DCT coefficient energy. Through calculating the average energy of each column of DCT coefficients of the different blurriness image to get vector Ceng, we realize that Ceng of different blurriness images almost has no difference on the head end, and has obvious difference on the tail end. According to this rule, no reference image quality assessment algorithm is proposed. The experimental results show that it is superior and its performance is highly correlated with people’s subjective feeling.