Study On Image Inpainting Algorithm Based On Local Structure

As one of the important research contents in image processing task, inpainting techniques generate image regions from the visually plausible backgrounds.Developing with the computer science and technology, the under maturing technology has been broadly applied in many fields: cultural relics protect, film special effects product, virtual reality, objects removal and so on. Image inpainting can be divided into two classes: methods based on variation partial differential equation and methods based on texture synthesis. The former establishes inpainting model using the known information and repairs damage along the illumination. The latter, based on image texture feature, populates the damaged area according to texture synthesis.Sample based image inpainting algorithm is a wide concern restoration model through the way of information copy. However in the actual study, it has shortcomings also, such as: unreasonable design of the priority setting, high time complexity, error repair, etc. by combining the structure tensor theory and the sample based inpainting algorithm, this paper presents a novel algorithm and realizes gray and color image recovery. The details and innovations are as follows:(1) Relevant theories of image inpainting are introduced in this paper. The representative models of inpainting algorithms based on partial differential equation and algorithms based on texture synthesis are analyzed for their advantages and disadvantages. And the sample based image inpainting algorithm is discussed in details.(2) We analyze the problems of the structure fracture caused by the inadequate image structure information proportion and the error inpainting caused by the inaccurate matching criterion in sample based algorithm. By introducing the structure tensor, a novel image inapainting algorithm based on structure tensor(IIST) is proposed. In this new algorithm, firstly, a structure control function is constructed to depict different regional characteristics of the image, and we also make a combination with the priority formula which increases the structure proportion in the repair process.Secondly, a new search matching relation equation and novel matching criterion are presented. Finally, damage repair and target removal simulation experiments have verified the effectiveness of the new algorithm on gray image inpainting problem.(3) The representative model of color image restoration is introduced, and traits of color spaces are discussed. The proposed IIST algorithm based on gray image restoration is extended for the color image inpainting. By converting the color space from RGB Lab, the extended model achieves color image inpainting by sub-channel processing. Experiments verify that the extended IIST algorithm based on the gray image can applied to color image inpainting as well.