Research On Non-reference Image Quality Assessment Algorithm Based On PCANet

No Reference Image Quality Assessment(NR-IQA)is the current research hotspot and future development direction of image quality assessment technology.It is important in image processing,robotics,machine vision,medical imaging and other fields.Application value.With the development of deep learning,deep neural networks are used by more and more researchers in the field of image quality evaluation with their powerful modeling and analysis capabilities.However,when deep neural networks are applied to image quality evaluation research,deep neural networks The convolutional neural network model also brings problems such as complex training,high requirements for parameter tuning skills and experience,and difficult theoretical analysis.At the same time,for the small data volume environment of the image quality evaluation data set,over-fitting is easy to occur.In response to the above problems,through the PCANet(principal component analysis network)research,in-depth research on PCANet-based non-reference image quality evaluation technology.First,this paper proposes a non-reference image quality evaluation algorithm based on PCANet.Perform deep feature extraction of the distorted image through the PCANet network,and then use the extracted features as input samples to train through Support Vector Machine(SVM)classifiers and Support Vector Regression(SVR)regressors Get the algorithm model.Experiments show that the algorithm can distinguish the types of image distortion and evaluate the image quality.The experiment was carried out on image quality evaluation databases such as LIVE,TID2013,and compared with some of the more commonly used non-reference image quality evaluation algorithms.The results show that the algorithm is better than Convolutional Neural Networks(CNN)and other algorithm indicators.Robust performance is better.Secondly,this paper proposes an image quality evaluation algorithm based on PCANet and Gabor wavelet.The first step of the algorithm uses Gabor wavelet to extract features from the image,and then the extracted features are further extracted through the PCANet network.Finally,the features are used as the input image and trained by the SVM classifier to obtain the algorithm model.Experiments show that the algorithm can distinguish the types of image distortion and rate the image quality.The experiments are carried out on image quality evaluation data sets such as LIVE and TID2013,and compared with the more commonly used image quality rating algorithms.The results show that the algorithm can achieve the effect of subjective evaluation and meet the characteristics of the human eye system.Finally,to sum up,the two algorithms proposed in this paper can achieve the subjective consistency with human eyes,and identify the type of distortion.In the future,we hope to improve the correlation filter of PCANet algorithm or combine with other DCT algorithms to improve the accuracy of the model and the efficiency of the algorithm.