Research On Exemplar-based Image Inpainting Algorithm

Image inpainting is a kind of image restoration technology to restore the damaged,distorted or defective images,it has very important theoretical and application value in the field of digital image processing and computer graphics,it aims to restore the damaged parts of an image with the rest and known image information and to make the observer could not see the image has been damaged or has been repaired.The current image inpainting technology can be divided into two categories.One is PDE-based inpainting algorithm,which can achieve a better result when the damaged area with small scale and low resolution.Another is texture synthesis-based inpainting algorithm.If images have rich texture information and large scale damaged area,this algorithm can achieve a better result.Criminisi algorithm is the typical algorithm in texture synthesis-based inpainting algorithms.After detailed analysis and in-depth research of the Criminisi algorithm and its improved algorithm,the improved algorithm has several shortcomings such as the unreliability of filling order,the time-consuming and the fixed sample block etc.,a new exemplar-based image inpainting algorithm is proposed.Firstly,combining previous improvements on Criminisi algorithm,this thesis changes the priority calculation formula from the sum of confidence and data to the product of them by introducing weight coefficients to avoid the data or confidence to 0 leading to inaccurate results,which makes the filling order more reasonable.Secondly,according to the local similarity characteristics of the image,a new set of search rules about the optimum matching patch is proposed in this paper.Test results show that the improved method not only can obviously improve the subjective quality of inpainting image,but also significantly reduce the time-consuming.Lastly,for the images with rich structure information and less texture information,using fixed sample size and local search window size is easy to produce the error of mismatch.Therefore,this thesis puts forward a set of dynamic adjustment strategy of exemplar and local search window to reduce error derivative and block effect late in the process,so as to improve the effect of inpainting.Simulation results show that the proposed algorithm is simple and easy to understand and implement.The process is fast and the results are satisfied.Therefore,the proposed algorithm has considerable practical value.