Research On No-reference Image Quality Assessment Model Based On Statistical Features

With the rapid development of science and technology,multimedia devices are widely used,and images as a digital resource with rich information have penetrated into people's daily life.There are a large number of digital images uploaded over Internet every day.Then,how to quickly find the image resources required by human beings from the massive digital images has become a hot spot of exploration.As an advanced visual information perception system,the human visual system(HVS)can efficiently acquire the key information in images when the quality of digital images is degraded in the process of image acquisition,storage and transmission.This study is devoted to the study of image visual quality evaluation by exploring the perceptual characteristics of HVS.Subjective evaluation is the most reliable quality evaluation method,but it is time-consuming and costly.In addition,it cannot be applied to practical applications.This paper proposes effective objective quality assessment methods by extracting perceptual features of images based on the human visual perception characteristics.The specific research content of this paper includes the following two aspects.(1)According to the human visual perception characteristic,the HVS first perceives the global information in the image when perceiving an image,and then specific local regions in the image.We propose a no-reference(NR)tone-mapped image quality assessment model based on local and global features.In the process of image viewing,the color and contrast change due to the introduction of noise,while the human eye can unconsciously perceive the unnaturalness of the image.Thus,we extract the statistical color feature,moment feature and entropy feature of the image to measure the unnaturalness of the image.At the same time,due to the limitation of the dynamic range,the tone mapped image cannot retain the structural information in the original high dynamic range(HDR)image.This paper considers the effects of different distortion types on structural features,using gray level co-occurrence matrix descriptors and zero-crossing to measure the changes of structure information of tone-mapped images.Finally,we use the support vector regression(SVR)to learning the mapping function from visual features to quality scores.The experimental results show that quality scores predicted by the proposed objective quality assessment model are highly consistent with the human subjective scores.(2)Considering the influence of changes between adjacent pixels on visual perception in the image,a perceptual quality evaluation algorithm based on spatial continuity is proposed in this paper.The proposed algorithm considers two types of spatial continuity: color continuity and texture continuity.First,we calculate four color descriptor maps from the distorted image(zero-order and first-order invariant color descriptor maps),and extract statistical color features based on four color descriptor maps to measure the color distortion of a distorted image;at the same time,considering the effect of texture variation on the visual quality of a distorted image,we use the local ternary pattern(LTP)descriptor to extract texture features of the distorted image.Then,we characterize the statistical color features and statistical texture features of the distorted image in the form of histograms,respectively.Finally,the statistical color features and the texture features are connected,and the final feature vector is input into the support vector regression(SVR)model to calculate the visual quality score of the distorted image.The experimental results show that the proposed perceptual quality assessment algorithm for the screen content images(SCIs)can effectively predict the visual quality scores of the distorted SCIs.This research covers visual perception feature extraction,visual quality score prediction,and NR quality assessment model construction,which effectively improves the prediction accuracy and learning rate of the NR IQA model.