Research On Image Quality Evaluation Based On Visual Characteristics

With the rapid development of digital multimedia technology,people’s demand for the clarity of images and videos has gradually increased.At present,image quality evaluation has been widely used in aerospace,medicine,security,transportation and other fields.Images and videos are prone to lose part of the information during the channel transmission,resulting in quality loss.Therefore,the accurate evaluation of image and video quality has application value.In this paper,the evaluation and performance analysis of stereo image quality are carried out from two aspects: non-reference quality evaluation and full-reference quality evaluation,and the human visual characteristics have been combined to evaluate and analyze the quality of stereo images,which mainly includes the following three aspects:(1)According to the characteristics of human vision,a method of stereo image synthesis based on Laplacian pyramid model has been proposed.Firstly,each layer of the Laplace pyramid contains different frequency bands of the image,and the stereo image pairs have been decomposed into pyramid sequences by using its multi-scale characteristics.Secondly,according to the theory of binocular weighted model,the fusion coefficients in different frequency bands have been analyzed and calculated to obtain the fusion sequence.Finally,by means of up-sampling and gaussian convolution of the fusion sequence,the bottom image has been reconstructed as the composite image of left and right viewpoints.(2)A quality evaluation of non-referenced stereo image based on feature extraction has been proposed.Firstly,on the basis of Laplace pyramid synthesis model,the frequency domain features and spatial texture features in each direction and frequency sub-range have been extracted according to the characteristics of spatial frequency multi-channel and texture sensitivity.Secondly,based on the support vector regression model,the relation mapping between feature set and subjective evaluation value has been obtained.Finally,the evaluation results of different kernel functions,penalty factors and kernel coefficients have been analyzed respectively to obtain the optimal parameters.(3)A full-reference stereo image quality evaluation based on image content has been proposed.Firstly,the weight distribution of image frequency domain information has been calculated based on the contrast sensitivity of HVS.Secondly,the edge information of the image has been extracted using the visual attention mechanism of HVS,and the frequency domain fusion information of the image has been calculated by using this region,so that the weight value of the high frequency information in the image is greater than the low frequency part.Finally,based on the full-reference stereo image quality evaluation theory,this method has been applied to the quality evaluation of monitoring images and videos,and the evaluation performance under different monocular coefficients and binocular coefficients has been analyzed to obtain the optimal linear weighting model.