Research On No-reference Omnidirectional Image Quality Assessment Algorithm Based On Multidimensional Features

With the rapid development of information technology,the emergence of smart phones and various high-pixel photo shooting equipment has promoted the continuous progress of virtual reality technology,and omnidirectional images have been gradually known and used by people.Through special viewing devices,omnidirectional images can provide viewers with a 360° immersive viewing field,enabling viewers to obtain an immersive viewing experience,which is favored by the majority of users.At present,omnidirectional images are widely used in all walks of life,such as medicine,education,industry and other fields.Although the omnidirectional image has many advantages,but like other ordinary image,the image of the film,the process such as transmission,compression and display inevitable will introduce some distortion,the distortion can't lead to image to show the most perfect state to the user,such as watching the phenomenon of fuzzy or ghosting,when greatly reduces the user viewing experience.Most of the existing image quality evaluation algorithms are proposed for planar 2-D(2D)images,which are difficult to be directly applied in the field of omnidirectional images.At present,there are few studies on the quality evaluation algorithms of omnidirectional images,and the existing algorithms do not fully consider the visual characteristics of human eyes,leading to the low performance of the proposed quality evaluation algorithm.In view of these reasons,it is necessary to study the quality evaluation of omnidirectional image.Combined with the current mainstream two types of omnidirectional image acquisition methods: direct sampling and down sampling,this paper focuses on the analysis of the image characteristics of the two types of omnidirectional image data sets,and studies the omnidirectional image quality evaluation methods based on the multidimensional perception characteristics of human vision.Firstly,combined with the characteristics of direct sampling omnidirectional image data set,this paper proposes the omnidirectional image structure characteristics calculated weighted gradient LPQ(GLPQ)on gradient graph by using image gradient and texture information.At the same time,the image entropy information,color information and natural scene statistics information were extracted as natural features.Through the effective combination of natural features and structural features,the objective quality evaluation method of omnidirectional image based on direct sampling data set was established.In addition,this paper designs a quality evaluation algorithm for the downsampling data set of image resolution adjustment for Head Mount Display(HMD).Considering that the image structure is still an important image information,this paper uses the combination of Gaussian derivative and LBP(Local Binary Pattern)texture operator to reflect the image structure information.Different from most quality evaluation methods,this paper adopts a variety of image gradient features in order to reflect the image quality comprehensively.Finally,in order to conform to the viewing characteristics of the human visual system,this paper adds the salient features of the image.A omnidirectional image quality evaluation method based on downsampling data set is established by combining multiple features.In order to verify the performance of the algorithm,the proposed data set on multiple omnidirectional image has carried on the experimental verification,the experimental results in this paper,the surface of the two types of omnidirectional image quality evaluation model on the two types of data sets has obtained the good effect,the evaluation results with the human visual system has a high degree of consistency,can effectively evaluate the quality of the omnidirectional image.