A blind quality assessment method for images using shape consistency feature

A new blind quality assessment method for images is proposed to solve the problem that the evaluation performances of learning based blind image quality assessment (BIQA) methods are sensitive to the contents of training samples and learning strategies. The method uses the shape consistency of conditional histogram based BIQA metric (SCCH) and does not need training and learning. The method calculates the joint conditional histograms of neighboring divisive normalization transform coefficients in distorted images, and then extracts shape consistency features from the histograms. Then the feature vectors are decomposed by scale, and a feature characteristic-subjective score dictionary is constructed by using public database. The lengths of the extracted features in the dictionary are compared with that of the distorted image and are sorted, and an interpolation using the subjective scores in the dictionary is then performed to calculate the quality score of the distorted image. Experimental results in two public databases show that the linear correlation coefficient between the SCCH and the image quality subjective scores of distorted images is more than 82%, and maintains a relatively high level. Compared with traditional BIQA methods, the SCCH has the following features, it does not need training, its quality score formula is simple, and the quality assessment system is easy to implement.