Evaluation of Perceptual Quality for Watermarked Images Based on Combination of Fuzzy Similarity Measures Using Neural Network

In this paper, focus is placed on the design a new evaluator to assess the quality of the image watermarking techniques. The main idea is the introduction of an image quality evaluator dependent on a combination of five fuzzy logic-based similarity measures and neural network. In the first stage, fuzzy similarity measures are computed as features of each pair of original and watermarked images and these features are used as input to neural network. In the second stage, these features are combined by using neural network to predict a subjective image quality, known as the Mean Opinion Score (MOS) automatically. The performance of the suggested evaluator is assessed in terms of good correlation with the MOS using the image watermarked database (IVC image database). Experimental results, using 210 tested images, show that the evaluation outputs correlate highly with the MOS scores.