Research On Omnidirectional Image Quality Evaluation Method Based On Human Visual Characteristics

Nowadays,with the arrival of 5G era and the continuous improvement of virtual reality technology providing strong technical support for immersive user experience,panoramic image has gradually attracted wide attention of researchers.With the birth of the concept of metacomes,panoramic images have been pushed to the craze again.However,all kinds of noise and loss will occur in the acquisition,compression,transmission and storage of panoramic image,which will affect people’s viewing experience and even their understanding of the image.In order to improve the user experience,it has become an important research topic for scholars to measure the quality of panoramic image to get the optimization index.Typically,panoramic images are viewed by an observer through a headset.When the device is worn,the visual range of the observer has certain limitations.In addition,the attention mechanism of the Human Visual System(HVS),which is responsible for processing visual information,plays a key role in perceiving images,which improves the efficiency of the brain.Compared with traditional images,panoramic images have more complex visual information,so the study of visual attention mechanism is an indispensable part of measuring the quality of panoramic images.Therefore,this paper carries out research on the significance detection of panoramic images and the quality evaluation of panoramic images.The main research contents are as follows:1.Considering that the existing significance detection model does not consider the characteristics of panoramic images,a significance analysis model for panoramic images is proposed in this study.Superpixel technology is used to take the pixel group as the basic processing unit of panoramic images.Compared with the pixellevel method,this method can effectively reduce the redundant information in the image and provide a more useful spatial structure for significance detection.Then,based on the Gestalt theory,this paper makes use of the scale invariance of the topology structure of the Boolean graph,separates the foreground and rear view of the image by calculating the Boolean graph of different threshold values,detects the significance region of the panoramic image by using different foreground-rear information,normalizes it by L2 norm,and finally obtains the final Gaussian fuzzy graph by using the linear combination method.In order to fully consider the characteristics of panoramic images,an accurate analysis model for the saliency of panoramic images is established in this paper.2.Inspired by the hierarchical perception of HVS,visual quality degradation can be described as a hierarchical process,that is,the perception process of input visual signals is gradually transitioning from local details to global semantics.Therefore,in order to better perceive the quality of panoramic images,this paper simulates the perception process of HVS.Firstly,the first affected detail information was measured,and the high frequency component reflecting the detail information of the panoramic image was separated by using the two-dimensional discrete wavelet transform,and the detail features of the image were extracted by combining the gray co-occurrence matrix.At the same time,the significance information is obtained by the significance algorithm using the measure of information entropy,which is used to represent the important local loss in the panoramic image.Finally,the natural scene statistics method is used to analyze the representation of global information on the whole semantic of panoramic images.In addition,this paper also considers the correlation and dependence between different color channels,and expresses the color information in the panoramic image by calculating the cross-channel local binary mode.By combining different features of panoramic image,a comprehensive perception of panoramic image quality evaluation model is established.In order to verify the performance of the proposed algorithm,the proposed model was tested using Support Vector Regression(SVR)technique on a panoramic image dataset.By comparison with other models,the experimental results show that the proposed panoramic image saliency modeling model can effectively detect the saliency region of the panoramic image,and the proposed panoramic image quality evaluation model can accurately predict the quality score consistent with the visual perception on the two publicly available data sets.