Research On Visual Optimization Method For User Quality Of Experience Improvement

With the advent of the information age,multimedia represented by images and videos is constantly influencing and changing our way of life and entertainment,which greatly enrich the user's visual experience.At the same time,users have higher requirements for their visual quality in uploading and viewing images or videos.Therefore,visual quality assessment and optimization based on quality of experience(Qo E)has attracted more and more attention.This paper combines visual perception from corresponding assessment to optimization,and then from 2D planar image to 3D image to study the visual optimization method for user experience quality improvement.The main contents are as follows:(1)An image aesthetic assessment method based on deep learning is proposed.Aesthetics is a very complicated and subjective problem.In order to evaluate the aesthetics of natural images,a deep neural network is used to simulate the human visual system,and a corresponding assessment model is established.Specifically,through the convolution neural network,the aesthetic information of the image and the scene category information of the image are separately extracted by using two channels,and then the two channels are combined at the fifth convolution layer,and the features are abstracted through several fully connected layers.After that,output the aesthetic level of image(beautiful or unattractive).(2)A contrast adjustment method based on image retrieval is proposed.Image contrast is unbalanced due to subjective or objective factors during image capture,and improving image contrast is a good way to improve the quality of user experience.Based on the image retrieval technology,a good quality reference image with similar content to the image to be enhanced is acquired,and three enhancement modes are combined: context-free,context-sensitive and brightness-adjusted.The image combined with parameters obtained by using these three processing methods.The aesthetic features and image entropy are used to guide and solve the multi-criteria optimization problem,and the image quality is used as the constraint to finally obtain the relevant parameters and the enhanced image.(3)A method simultaneous object size and depth adjustment is proposed.In order to enhance the user's 3D stereoscopic experience,a method of user and image content interaction is proposed.By specifying the object or region of interest,the user can set the corresponding adjustment coefficient and adjust the object size and depth.The mesh deformation is used to control the feature points on the object to protect the shape of the object,adjust the size of the object,and maintain the comfort of the depth.The adjusted image is sized to achieve a predetermined effect,and its depth can be enhanced within a range of comfort.(4)A method of global and local adjustment is proposed to realize 3D image zoom.In order to further enhance the user's 3D stereoscopic experience,avoid the distortion of the 3D image caused by optical zoom,and adjust the image plane to realize the effect of the stereo camera moving on the real depth,thereby realizing stereo image zoom.After the user specifies the focus intensity,the two content adjustments are used for global adjustment and local adjustment,and the global adjustment is performed for global adjustment,and the local adjustment is performed for local area fine adjustment to ensure image quality.Eventually the zoom of the image content is achieved,and the depth of the 3D image will adaptively vary with the adjustment of the image.The four research methods proposed in this paper,from assessment to optimization,from 2D image form to 3D stereoscopic image form.Compared with other research methods,there are indicators or user experience improvements.