Blind Quality Assessment of Super-Resolution Images using Relative Gradient Statistics of Salient and Non-salient Objects
Malik Waqar Aziz, Imran Fareed Nizami, Muhammad Majid · 2021
In recent years, due to the exponential increase in the usage of multimedia content in our everyday lives, blind image quality assessment (BIQA) has gained a lot of importance. Distortions of many types may be introduced in the images due to limitations in modern technology that needs to be assessed. Blind assessment of super-resolution (SR) images has attracted many researchers to address the limitations in BIQA algorithms. In this paper, a new technique for blind quality assessment of SR images is proposed. We have used the visual saliency technique to extract salient and non-salient objects from the input SR images using you only look once (YOLO) deep learning model. The relative gradient orient features are extracted from salient and non-salient objects, which are then used to predict the score by using Ada-boosting back propagation neural network (Ada-Boosting BP NN). The overall score of the SR image is calculated by the weighted average of the scores obtained from salient and non-salient objects. We have evaluated the performance of the proposed method on three independent SR image databases i.e., MY, Quality assessment database for SRI (QADS), and super-resolution image quality assessment Javeriana (SRIJ). Experimental results signify that the proposed method shows better performance on all three SR databases.