Abstract:The mission of blind image quality evaluation is currently a challenging computer vision problem. Due to the shortage of reference images, it is hard for blind image quality evaluation methods to achieve the same performance as full reference image quality evaluation methods. In addition, current quality evaluation methods are difficult to effectively forecast the quality scores of synthesized distorted images as well as real distorted images. To address such issues, this study proposed a cyclic generative adversarial network composed of a quality perception network and a quality regression network on the grounds of generative adversarial networks. For further enhancing the predictive performance of quality aware networks, this study proposed using attention blocks for adaptively fusing high-level semantic features and low-level semantic features. It extracted content and distortion information from images through an image quality evaluation method on the grounds of content perception and distortion inference. And according to the different properties of the extracted features, adaptive fusion blocks were used for adaptively fusing content features and distortion features. Experiments showcased that the Spearman order correlation coefficient and Pearson linear correlation coefficient obtained by the proposed method on multiple datasets were higher than other similar methods. At the same time, the proposed method has achieved good prediction results on various types of distorted images, and has surpassed other methods. The prediction accuracy of the proposed method on five types of distortion was 0.971, 0.963, 0.984, 0.971, and 0.926, respectively. The proposed method achieved the highest predictive accuracy on all distortion types in the LIVE dataset, with predicted accuracy values of 0.973, 0.965, 0.984, 0.963, and 0.944, respectively. In summary, the proposed method not only achieved good prediction accuracy, but also had strong generalization performance in cross dataset testing. This provides a scientific and effective research direction for blind image quality evaluation.

Blind Image Quality Assessment Using a Deep Bilinear Convolutional Neural Network

Deep Blind Image Quality Assessment Using Dynamic Neural Model with Dual-order Statistics

On the Use of Deep Learning for Blind Image Quality Assessment

Deep Superpixel-Based Network For Blind Image Quality Assessment

Blind image quality assessment using Beltrami filter-based contrast features (BF-bCF) & LSTM network

Blind Image Quality Assessment With Active Inference

Blind Image Quality Assessment Using Multi-Stream Architecture with Spatial and Channel Attention

Learning from Synthetic Data for Opinion-free Blind Image Quality Assessment in the Wild

Blind Image Quality Assessment Via Cross-View Consistency

Deep neural network based distortion parameter estimation for blind quality measurement of stereoscopic images

Blind 3D Image Quality Assessment Based on Multi-scale Feature Learning

Image Quality Assessment Without Reference By Mixing Deep Learning-Based Features

Deep Ensembling for Perceptual Image Quality Assessment

Quality-aware Pre-trained Models for Blind Image Quality Assessment

Deep Neural Network for Blind Visual Quality Assessment of 4K Content

Blind Image Quality Evaluation Method Based on Cyclic Generative Adversarial Network

Blind Image Quality Assessment Based on Multichannel Feature Fusion and Label Transfer.

A Multiscale Approach to Deep Blind Image Quality Assessment

Blind Image Quality Assessment Using Semi-supervised Rectifier Networks

Blind Quality Assessment for in-the-Wild Images via Hierarchical Feature Fusion and Iterative Mixed Database Training

A shallow convolutional neural network for blind image sharpness assessment