Research On A New Method For Quality Assessment Of Unreferenced Images

Image quality assessment(IQA)aims to output a scalar termed quality score,which is closely related to the human vision system(HVS).No-Reference IQA(NR-IQA)is to build a quality regression model for the perception of HVS without any reference information.NR-IQA is essential for the field of image processing,which is mainly divided into two research directions:(1)image understanding,to explore the relationship between computer vision and HVS;(2)improving the baseline algorithm via quality information.This thesis focuses on these two directions to carry out a study of NR-IQA methods,the specific works are as follows:(1)Proposed a novel NR-IQA method based on refined natural scene statistics(NBIQA).The proposed NBIQA first investigates the performance of a large number of candidate features from both the spatial and transform domains.Based on the investigation,we construct a refined natural scene statistics(NSS)model by selecting competitive features from existing NSS models and adding three new features.Then the refined NSS is fed into the SVM tool to learn a simple regression model.Finally,the trained regression model is used to predict the scalar quality score of the image.Experimental results show that the proposed NBIQA performs better than current mainstream IQA methods in terms of synthetic and authentic image distortion.(2)Proposed a controllable list-wise rank learning-based NR-IQA method(CLRIQA).To be specific,CLRIQA first present an imaging-heuristic approach,in which the over-underexposure is formulated as an inverse of the Weber-Fechner law,and fusion strategy and compression are adopted,to simulate the authentic distortion and generate the rank image samples.These samples are label-free yet associated with quality ranking information.Then a controllable listwise ranking loss function is designed by setting an upper-lower bound of rank range and introducing an adaptive margin to tune rank interval.Finally,both the generated rank image samples and proposed CLR are used to pre-train a convolutional neural network.Moreover,to obtain a more accurate prediction model,CLRIQA take advantage of the IQA datasets to further fine-tune the pre-trained network.Various experiments are conducted on four IQA datasets,and experimental results demonstrate the effectiveness of the proposed CLRIQA method.(3)Proposed an unsupervised face image quality assessment with similarity distribution distance(SDD-FIQA).We argue that a high-quality face image should be similar to its intraclass samples and dissimilar to its inter-class samples.Thus,we propose a novel unsupervised face image quality assessment(FIQA)method that incorporates similarity distribution distance for face image quality assessment.SDD-FIQA generates quality pseudo-labels by calculating the Wasserstein distance between the intra-class and inter-class similarity distributions.With these quality pseudo-labels,we can be capable of training a regression network for quality prediction.Extensive experiments on benchmark datasets demonstrate that the proposed SDDFIQA surpasses the state-of-the-arts by an impressive margin.Meanwhile,our method shows good generalization across different recognition systems.