Image Inpainting Algorithm Based On Exemplar-Based Filling And Curvature Driven Diffusion Model

As a hot research field in digital image processing,image inpainting attracts the attention of lots of researchers at home and abroad.At the same time,image inpainting is also a basal field,and it is great significant to the development of digital image processing.The objective of image inpainting is to restore the damaged image information,keep its integrity,and not be perceived by the human.Basing on the intensive study of partial differential equation,wavelet transform and sample filling,this paper puts forward two image inpainting algorithms,the innovation points mainly in the follow two aspects:Firstly,a remote sensing image inpainting algorithm based on non-local exemplar-based filling and adaptive curvature driven model is proposed.Firstly,Euclidean distance threshold is introduced to non-local means algorithm to restore the image by non-local sample filling based on regularization.Secondly,Curvature Driven Diffusion model based on guide-function of gradient is proposed,the intensity of diffusion can be adjusted adaptively according to gradient and curvature.Lastly,the diffusion model is applied to the filled image.The new algorithm can avoid the cases instead of CDD model,such as the false edge in some extreme cases,the staircase effect and the slow diffusion velocity.Then,an image inpainting algorithm based on shock filter-based adaptive CDD model and wavelet decomposition sub-bands processed separately is proposed.Firstly,Euclidean distance and structure similarity are used to calculate the filling block by weight value.Secondly,the filled image is decomposed into low-frequency sub-band which represented the image's detail information and high-frequency sub-band represented the edge information by undecimated discrete wavelet transform.And according to the characteristics of the two sub-bands,the different methods are used.For low-frequency sub-band,a shock filter-based adaptive CDD model is proposed,the model can restore the image edge better.For high-frequency sub-band,sample filling based Euclidean distance and structure similarity is used.Lastly,reconstitute the restored wavelet sub-bands and obtain the restored image.A large number of simulation experiments verity the effectiveness of the two proposed algorithms.The first algorithm avoids the cases when gradient and curvature are both large in the process of restoring,such as the false edge,the staircase effect and the slow diffusion velocity.By this algorithm,the restore effect has been greatly improved.The second algorithm can preserve the edge information and the detail information while restoring the image effectively.