Research On Image Variational Quality Assessment And Segmentation Methods

Image processing using variational method or partial differential equation method is one of the main research directions in the field of computer vision.The method establishes the energy generalization function according to the defination of the problem,and obtains the partial differential equation that minimizes the energy function based on the variational method.The energy generalization usually consists of two parts: the rule term and the data fidelity term.The variational rule term can be used to carve image features,including image texture,shape,color and spatial relation features,etc.The data fidelity term has the functions of smoothing and edge preserving.The reasonable adaptive data fidelity term can deal with various problems in image restoration.Based on the variational theory,the variational image quality assessment method and the deep-sea image variational segmentation method are proposed.Quality assessment is widely used in image processing domain.The traditional quality assessment models can hardly express the visual perception of the human eye system,and the existing methods based on the variational theory only consider the edge as a visual feature,so the assessment results can not be consistent with the human subjective feelings.The deep-sea images generally suffer from serious noise and poor clarity due to light absorption and suspended particles.The existing image segmentation models have good segmentation effects,however,due to the limitation of data term,which fail in deep-sea image segmentation and result in unsatisfactory image segmentation results.In summary,the models of image quality assessment and the deep-sea image segmentation need to be improved.In view of the problems of image quality assessment and deep-sea image segmentation problems,on the basis of summarizing and analyzing the existing methods,the main research contents and innovations are as follows:(1)Based on the framework of structural similarity(SSIM)image quality assessment,a novel quality assessment model is proposed by fusing the varitional regularization features.Considering that the detailed features such as edge and texture of the image can be engraved by the variational rule term,the method extends the low-order term to the higher-order regularization term.The varitational high-order term is introduced into SSIM method,and an objective quality assessment model that can be consistent with subjective perception is established.(2)On the basis of the Mumford-Shah multiphase segmentation model,combining the dehazing characteristics of the dark channel prior(DCP)theory,a deep-sea image segmentation method is constructed.The data term for deep-sea image dehazing is provided based on the DCP model.Combining it with the Mumford-Shah multiphase segmentation model,we establish a variational segmentation model for deep-sea image.To improve the computational efficiency of the algorithm,the fast split Bregman projection and alternating direction method of multipliers are introduced for iterative solution of the energy function.(3)In order to verify the effectiveness of the proposed methods,image quality assessment and deep-sea image segmentation experiments are conducted.Experimental comparisons with traditional models are carried out on the LIVE,TID2008 and TID2013 image databases,verifying that the image quality assessment model based on the SSIM and variational regularization features is closer to the subjective evaluation scores of the human eye.The deep-sea image segmentation experiments are verified on Coral reef datasets such as Eilat,Coral-Net and Mosaics UCSD to demonstrate the dehazing and segmentation accuracy of the proposed segmentation model,as well as the efficiency and convergence of the algorithm.