Study On Digital Image Inpainting Based On Wavelet Transform And Harris Operator

As intuitive information, Images have been used in people’s daily life widely, but man-made or natural factors will make the image appear damaged which affect their integrity. Digital image inpainting technology, also called image completion or filling the holes, refers to use some methods to fill in the missing regions (holes) in an image. Digital image inpainting techniques have been widely investigated in various applications of special effects, image restoration, image transmission and coding and so on.This paper focuses on the digital image inpainting method whose missing regions are already known and studies the exiting digital image inpainting techniques from three perspectives in depth. The main work and dedications of this paper are illustrated on the following three topics:(1) As for issue of the current image inpainting algorithm based on dictionary learning which couldn’t grasp the texture detail well, this paper presents an improved image inpainting algorithm based on multi-scale dictionary learning in the wavelet domain. Firstly, the subband dictionary used in our method is generated by incorporating the wavelet transform and KSVD dictionary learning approach together. Then combine each subband dictionary into a global one. Finally, this paper analysis the influences of wavelet basis and decomposition level in theory and points out the deficiency at last.(2) For the problem existing in the Criminisi algorithm, this paper proposes an adaptive image inpainting method based on the Harris corner operator. The method improves the Criminisi algorithm from three aspects respectively:the definition of priority, the selection of template size, the similarity matching strategy. Firstly, the method uses the Harris corner operator as the data term of priority and changes the combination form of the confidence term and the data term; Secondly, the method proposes an adaptive strategy to reduce repair failure caused by the fixed template. The adaptive strategy differentiates the corner, edge and flat region in an image according to the corresponding relationship between the two eigenvalues. Finally, the method adds the gradient term on the SSD term to construct a weighted similarity matching criteria during the block matching.(3) The diffusion-based method can only fix the small holes and the exemplar-based method can produce poor result for the uneven bright images. To solve the above two problems, this paper designs an image inpainting method based on wavelet transform. Firstly, FMM algorithm is used for image preprocessing. Then the image after preprocessing uses wavelet separation to get high frequency and low frequency part of image. Criminisi algorithm with the improved priority is applied on the texture part to repair. Then, obtain the final image inpainting results by combining the low and high frequencies together.The computer simulation results show that the methods this paper proposed can effective inpaint the degraded image and have a wide range of application. Compared with the current similar algorithms, our methods can achieve better results.