No-reference quality metric based on fuzzy neural network for subjective image watermarking evaluation

Abstract

A new no-reference image quality metric is proposed to estimate the quality of watermarked images automatically based on fuzzy neural network. The aim is to use fuzzy neural network to learn the highly nonlinear relationship between the fuzzy similarity measures and the subjective quality rating, known as the mean opinion score (MOS) obtained from human viewers. In fact, our metric consists of four stages: first, transform watermarked image (crisp data set) into a fuzzy watermarked image (fuzzy data set). Second, fuzzy filtering process is applied to fuzzy watermarked image in order to generate its filtered image. Third, we use fuzzy watermarked image and its filtered image in the calculation of the fuzzy similarity measures as input to a neural network. Fourth; these measures are mixed using neural network model. The output of proposed metric is a single value corresponding to the MOS scores. Experimental results show that fusion of fuzzy similarity measures through the neural network indeed is able to accurately predict perceived quality of watermarked images.