A Machine Learning Algorithm for No Reference Image Quality Assessment using Non-Subsampled Contourlet and Curvelet Transform

Image quality assessment(IQA) predicting the quality of an image has lots of application in the communication area and entertainment industry. The IQA without any reference image needs to be done in several practical situations. For instance, to measure the image quality after decompression, the reference image will not be available. In such cases, measuring the IQA without a reference image is a challenging problem. The proposed algorithm is experimentally found to be more efficient than well-known algorithms. We use the transforms such as Curvelet Transform and Non-Subsampled Contourlet Trans- form(NSCT) and we fit the coefficients of NSCT in Asymmetric Generalized Gaussian Distribution(AGGD). We found that this fit is a better fit among several distributions that we explored. Support Vector Regression(SVR) is used for finding the quality measure of an image and SVR is using features derived out of AGGD and Curvelet Transform coefficients. The performance of the proposed algorithm is experimentally found to be efficient on the standard database.