The Study Of Digital Image Completion Algorithms

As a research hotspot in the field of digital image processing and computer vision,the digital image completion is to fill the data missing region,keep the continuity of the structure and texture,and then guarantee a natural and reasonable completion result.Digital image completion method is mainly comprised of two categories: inpainting technique and completion technique.The former one aims to deal with smaller damaged region,such as scratch,spot or words removal.It is mainly processed by the diffusion-based algorithms.While the later one is used to complete larger damaged region or fill the data missing region after removing larger target.It is mainly processed by the exemplar-based algorithms.C urrently,the image inpainting technology is relatively mature,so it is not focused in this thesis.However,since current researches mainly focus on the image completion technology,it is discussed in detail in the thesis.Firstly,this thesis presents the digital image completion's background and research status,as well as its significance.Moreover,the exemplar-based texture synthesis algorithms are analyzed in detail,as well as some important technologies.In addition,this thesis introduces some improved exemplar-based algorithms,such as robust priority function,and several optimal patch searching methods,such as least mean square deviation methods,isometric transformation method,multiresolution completion method and optimal matching patch sparse representation method.Secondly,this thesis introduces an image completion a lgorithm based on curve fitting.The color image decomposition method is introduced to obtain the structural image and texture image.Then,the damaged structure can be fitted with Bezier curve fitting method.It is benefit to keep the continuity of image structure by using the fitted curve to guide the completion direction,as well as improve the completion result.In addition,the Evolutionary algorithm is used to obtain the optimal patches and decrease mismatching.By us ing the completion process of structural image to guide that of the texture image,the proposed algorithm can guarantee the perfect effect of both structure and texture components.Experimental results show that the proposed method can achieve perfect effect for images with damaged structure.Finally,the thesis presents a robust image completion algorithm based on isophote detection.By using the Guided filtering to remove texture detail and noise,the structural image is obtained.Then,the traditional isophote function is improved with the edge stopping function in anisotropic diffusion approach,and combined with the traditional priority function.Using the isophote function to guide the texture completion process,the proposed algorithm can obtain a robust completion effect.Experimental results show that the proposed algorithm can preserve image structure effectively,and obtain relatively ideal effect in the aspect of texture synthesis with il umination changes.