No-reference blur image quality assessment based on Simulated Annealing and General Regression Neural Network

In order to improve the accuracy and efficiency of no-reference blur image quality assessment based on General Regression Neural Network. We choose Simulated Annealing algorithm to optimize the method. Using LIVE (Laboratory for Image & Video Engineering) database as the initial study database. 174 images from LIVE database are assigned randomly to two groups. Phase-matched images generated by phase transformation. We can get Gray Level Co-occurrence Matrix form phase-matched images. Then, get the energy, Entropy, correlation, contrast and homogeneity of these five characteristics indexes. Using the above indicators as input data and using Difference Mean Opinion Score as output data. Training neural network model on this way. In order to improve the accuracy and efficiency, using the Simulated Annealing algorithm to find the optimal smoothing factor parameter. Finally, spearman correlation coefficient of objective and subjective data is 0.9319. Pearson correlation coefficient of objective and subjective data is 0.9328. The results show that, this algorithm fits Difference Mean Opinion Score well. It predict better on image quality assessment.