Convolution Neural Networks Based Blind Quality Predictor
Prudhivi Anuradha, K. Rajkumar, J. Ravichander, Ch Padmaja · IOP Conference Series Materials Science and Engineering · 2020
Abstract Image recognition focused on convolution neural networks (CNN) in various areas of computer vision and image processing has recently proved to deliver state of the art output. Nonetheless, applying deep CNN to an NR-IQA remains a long-standing challenge owing to crucial challenges, i.e. the absence of a testing framework. The CNN test has not yet been performed. In this paper suggested a NR-IQA system focused on CNN which can solve the problem effectively. The approach suggested the DIQA divides the NR-IQA into two stages: 1 analytical portion of the illusion, and 2 one subjective component of the visual system. In the first step, the CNN learns to forecast the objective error map and then in the second phase the algorithm is able to predict subjective values. In addition, we are introducing a reliability chart to supplement the inaccuracy of the objective error map in the homogenous area. To further improve the accuracy, two simple handmade features were used. Therefore, we suggest a way to show graphs of perception errors and examine what the deep CNN model has experienced. The DIQA received the most sophisticated precision on the different databases in the experiments.