Image Quality Assessment by Combining Fuzzy Similarity Measures using Neural Network

In this paper, we focus on the design a new evaluator for image quality evaluation. The main idea is the introduction an image quality evaluator dependent on combine five of fuzzy logicbased similarity measures (S1, S2, S3, S4 and S5) using neural network. In first stage, compute fuzzy similarity measures as features for each pair of original and degraded images and use these features as input for neural network. In second stage, combine these features by using neural network to predict Mean Opinion Score (MOS) automatically. The performance of the proposed evaluator is evaluated in terms of good correlation with the MOS using the IVC image database. Experimental results, using 210 tested images, show that the evaluation outputs correlate highly with the MOS scores. The predicted MOS values have a linear Pearson correlation coefficient of 0.996 and Spearman ranked correlation of 0.987. These results compare very favorably with the results obtained with other methods. Keywords-image quality; fuzzy logic; fuzzy similarity measures; neural network.