The Theory And Application Of Perceived Variational Model Algorithm

As an important way for human perception information,images are becoming more and more important in this information age.And with the development of science and technology,people are not only satisfied with seeing ordinary images,but also seek to see clear images.However,due to the influence of various factors,certain degradation will occur in the process of capturing,transmitting and receiving images,which will obscure some information in the images,that is not conducive to people's observation and subsequent processing needs.Therefore,some algorithmic techniques are needed to further enhance and restore the images.The purpose of image enhancement is to enhance certain effective information and details in the image according to specific requirements,restrain some useless information,and improve the visual effect of the processed image.Considering the final processing results of the images is judged by the human eyes,in order to better enhance the images,this paper analyzes and studies the characteristics of the human visual system to improve the original image enhancement algorithm,then propose a better enhanced model that can better simulate the characteristics of the human visual system and applied to the color cast image and low contrast image zooming,image dehazing and low light image enhancement to verify its effectiveness.The main work of this thesis includes the following two aspects:Correct the color cast,increase the contrast and image zooming are performed separately in traditional algorithms for color cast image and low contrast image zooming,to simplify the process,a variational model for color cast image and low contrast image zooming which can compound three processes together is proposed.Firstly,the three contrast functions in the visual perception enhancement variational algorithm are analyzed and a new contrast enhancement term is defined.Then,the color cast correction term and contrast enhancement term are introduced into the image zooming variational model.Finally,the optimization solution is obtained by using the gradient descent flow method minimize model.The experimental results show that for the color cast image and low contrast image zooming problem,this model can effectively improve the image clarity and protect the image detail information.To solve the problems that the low contrast and low saturation in fog images,an image dehazing variational model based on visual perception is proposed.This model makes two improvements to the variational enhanced model based on visual perception.On the one hand,the gray-world hypothesis is corrected by estimating the average value of the clear images based on the atmospheric scattering model.On the other hand,the color saturation of the image is increased by maximizing the contrast between the channels in the energy term.The experimental results show that the proposed image dehazing variational model based on visual perception is superior to other dehazing methods both qualitatively and quantitatively.Especially,when the light source is not uniform,the method in this paper can restore the color of the image better and avoid strong color artifacts.The low light images have extremely low contrast and strong noise.To solve the problem,an adaptive variational model for low light image enhancement is proposed.Firstly,the transmission of the low light image after inversion is roughly estimated based on the luminance component,and the transmission is further detailed by using the Retinex algorithm in order to retain more image details.Then,in order to suppress the amplification of the noise and keep the information of edges,the regularization parameters are adaptively adjusted according to the priori of the bright channel and the local variance.Finally,the optimal solution of energy functional equation including transmittance and restored image is obtained by alternating iterative optimization method.Experimental results show that the proposed model can effectively enhanced low light image,retain more details,and suppress noise amplification.Compared with ₁l-norm regularization method,the larger the image size,this model has higher computational efficiency and more obvious advantage of computing time.