Study On Image Quality Assessment Algorithm And Its Application In Printing Quality Evaluation

China is a big country of printing import and export.Printing,as an important carrier of information transmission,its image quality directly affects the sufficiency and accuracy of information expression.With the development of artificial intelligence,printing image quality assessment(IQA)method based on computer processing technology has attracted more and more attention because of its high precision and high automation level.The image quality assessment method simulates human’s perception process of image quality and constructs an evaluation algorithm as consistent as possible with the subjective scoring results.The key lies in the cognition of the perception mechanism of human visual system(HVS).Aiming at the problem of low consistency between existing algorithm scores and subjective scores,this dissertation deeply explores the HVS perception characteristics related to image quality,and constructs new image quality assessment algorithms from the perspectives of nonlocality,combination of global and local,visual significance and nonlinearity,so as to further improve the consistency between algorithm scores and subjective scores.At the same time,combined with the proposed algorithms and the types of print quality distortion,print quality evaluation algorithms are constructed.The research work of this dissertation includes the following five aspects:(1)An image quality assessment algorithm based on non-local gradient is proposed.Most of the existing IQA algorithms are designed based on local structure similarity,but human’s subjective perception of images is a high-level and semantic process,and semantic information is essentially non-local.The image quality evaluation should consider the non-local information of the image.This dissertation breaks through the classic algorithm framework based on local information,proposes an algorithm framework based on non-local information,and constructs an image quality evaluation algorithm based on non-local gradient within this framework.The algorithm measures the non-local gradients similarity between the reference image and the distorted image,then the similarity is used to predict the image quality.The experimental results show that the algorithm improves the prediction performance of compressed distorted image quality.(2)An image quality assessment algorithm combining global and local changes is proposed.As image information is presented by the change in intensity values in the spatial domain,the gradient,as a basic tool for measuring the change,is widely used in IQA models.However,gradient can only measure local changes,while when HVS perceives an image,it can perceive both local changes and global changes.Based on this characteristic of HVS,this dissertation proposes an image quality assessment algorithm that combines global and local changes.The algorithm uses fractional derivatives to measure the global changes and uses the gradient magnitude to measure the local changes of the image.Synthesizing the two aspect changes,similarity map between reference image and distorted image is calculated,and then objective score of the image is obtained.The experimental results show that the algorithm can more accurately simulate the perception process of HVS on image quality,and the objective scores given on the transmission distorted images have better consistency with the subjective scores.(3)A visual saliency-guided edge-strength similarity algorithm is proposed.Classical visual saliency-based IQA algorithms use visual saliency in the pooling stage,emphasizing that regions with strong saliency contribute more to the overall image quality,and regions with weak saliency contribute little to the overall image quality.Different from the classic method,this dissertation proposes the use of visual saliency in the local image quality measurement stage which uses the visual saliency to adjust the calculation of local image quality adaptively,emphasizing that the local quality degradation perceived by HVS is jointly determined by objective degradation degree and visual saliency.A visual saliency-guided local image quality measurement framework is proposed,and within this framework,the edge-strength similarity algorithm is extended,and a visual saliency-guided edge-strength similarity algorithm is proposed.The experimental results show that the algorithm improves the prediction accuracy of noise distorted image quality.(4)Two image quality assessment algorithms based on the nonlinear characteristics of HVS are proposed.Firstly,the activation function is very important in the deep learning method which ensure the nonlinear approximation ability of the network.Inspired by the deep learning method,this dissertation proposes an evaluation algorithm based on the similarity of activation edge-strength,which uses the activation function to perform nonlinear processing on edge-strength and non-linear processing on image local quality calculation.Secondly,human evaluation of image quality is carried out on the HVS perception space.In this dissertation,the ability of HVS to perceive changes is affected by the upper threshold,and an image quality evaluation algorithm with adaptive truncated gradient is proposed.In the algorithm,the gradient magnitude undergoes non-linear processing,the part less than the upper threshold is retained,and the part greater than the upper threshold is truncated.The experimental results show that the proposed algorithms are competitive in blurred image evaluation performance compared with the current algorithms.(5)Combined with the proposed image quality assessment algorithms and the types of printing distortion,printing quality evaluation algorithms are constructed.Firstly,considering the influence of color distortion on image quality measurement,a color image evaluation algorithm based on color and structural distortion is constructed,and then considering the color perception characteristics of human vision,a color evaluation algorithm based on color quantization is constructed.The effectiveness of the algorithm is verified on the printed test images.Secondly,aiming at the distortion of image gamut mapping caused by different color gamut in the process of printing the image on the substrate,a gamut mapping image quality evaluation index is constructed.The experiment results on the gamut mapping images show the effectiveness of the algorithm.Lastly,aiming at the problem of color image output quality in monochrome printing equipment,a color-tograyscale image quality evaluation index is constructed to measure the fidelity of the information before and after conversion.The performance of the algorithm is verified through the accuracy and preference experiments on the grayscale images.On the one hand,the research results of this dissertation reveal the importance of HVS non-local,global and local combination,visual saliency,and nonlinearity in the perception of image quality.On the other hand,it provides a theoretical basis for the research of HVS perception mechanism to a certain extent.The algorithms proposed in this dissertation improve the prediction accuracy of image quality,and provide the possibility for the application of IQA methods in engineering practice,especially for the application research of IQA algorithms in the printing field,which can provide technical support for automatic control and manage printing quality.