Research On Image Inpainting Algorithm Based On Structure Group Sparse Representation

Making reasonable conjecture and filling damaged areas of an image by certain rules according to the existing information is called image inpainting,which aims to make the reconstructed image close to the original image maximally or achieve a natural visual effect.Damaged areas of images differ in shape under various application scenarios.Different types of repairing algorithms are proposed correspondingly.In this thesis,the classic algorithms are studied,and their advantages along with the disadvantages are discussed.Then an improved structure group sparse representation(SGSR)model is proposed on this basis.The details are as follows:The SGSR image inpainting algorithm proposes the concept of structure group and uses singular value decomposition(SVD)on the estimate of a structural group to obtain the dictionary,then utilizes split Bregman iteration(SBI)algorithm to solve the optimization model to get sparse coefficients,and finally adopts the dictionary and the sparse coefficients to repair an image.In some extent,this algorithm solves the problem that the traditional sparse representation algorithm ignores the similarity between image patches,which will result in the fact that structures and textures in a reconstructed image are not natural enough.As the bilinear interpolation(BI)algorithm is used to calculate the estimate of a structural group,the SGSR algorithm does not fix the missing patch well.In this thesis,in order to get the estimate of a structural group more accurately,we exploit the Curvature Driven Diffusions(CDD)model and Criminisi algorithm to evaluate images with small missing area and images with big missing area respectively.Experimental results and data show that the proposed algorithm performs better on subjective vision and objective indexs than many algorithms,including the SGSR algorithm.Traditional image inpainting algorithms have limited information on the source area of a single image,making it difficult to repair images with complex structural textures and large missing areas.In order to solve this problem,this thesis proposes an image inpainting algorithm based on multi-view structure group sparse representation.First initialize and register multiple reference view images to the same perspective as the target image,then use MRF to combine the known information on all registered reference view images to fill the missing areas of the target image.Next,Poisson fusion technique is used to eliminate the inconsistency of the photometric edge.Finally,for the remaining holes,a strategy is proposed to select the most adaptive reference view image,which helps repairing the target image using a multi-view-based structure group sparse representation model.Experimental results show that the proposed algorithm has better effects on large-area missing regions than other algorithms.