Research On Image Inpainting Technology Using HSI Color Space And Its Application

Digital image process grows its importance with the development of computer technology and multi-media technology. Image inpainting is one of the most important area of digital image process. It is defined as some method that fills the missing area of an image using known part of this image with some algorithm. Its purpose is to inpaint the broken image acceptable to human eye vision. With the image inpainting algorithms development, its application becomes more widely. Image inpainting now is widely used on cultural relic repair and research, object removal in image, procedural landscape generation, data encryption, etc.There are two kinds of image inpainting methods which are most widely used. The first kind is inpainting method based on partial diffusion equation (PDE). This kind of method performs well on small broken area inpainting. It inpaints image in pixels by using the thermal diffusion principle of physics. It diffusions the known information from area near the broken part. But when inpainting large area or rich structural information this method will lead some fuzzy distortion. The second method which is based on texture synthesis performs better than the former method for this kind of broken images. This kind of method inapaints images by pixel blocks. It synthesis the image texture that in line with the visual continuity in the broken part using the known part of the image and driven by isophotes. This kind of method was proposed by Criminisi first. The two key factors of this method is the calculation of inpainting priority and choose of sample block of texture synthesis. The improvement of this method are mostly on these two aspects.The main innovation of this thesis is as below:(1) An improved method is proposed based on the Criminisi method. Gray scale gradient is replaced by HSI color space gradient to improve the inpainting priority calculation and the HSI color space gradient statistical information is also considered. To improve choose of sample block of texture synthesis, Local consistency of image is used to accelerate and the isomorphism geometric transformation is used to improve the inpainting quality. Some object removal and image inpainting simulation experiments are done and the results are compared with the results of Criminisi method, Wexler method and Guillemot method.(2) In this paper, one improved mean dual-scale edge structure similarity (MDESSIM) is proposed to quantized evaluate the comparison. The comparison of the experiment results shows that the methods proposed by this paper performs better in diffusing the structural information and cause less fuzzy distortion.(3) At the end of this paper, the application of this method in video inpainting and image compressing is studied. The image inpainting method is combined with one moving object extraction method and whole scene of attached background building method to inpainting videos that the camera is immovable or moved regularly. The image inpainting method is combined with image edge detection and image dilation to compress/decompress an image. The size of dilation window is regulated dynamically. This method can compress/decompress image in a good compression ratio and keep the image quality.Aiming at big-scaled broken or rich-structured broken image, some inpainting method using HSI color space is proposed in this thesis, the mean dual-scale edge structural similarity is improved to evaluate its quality. Video inpainting method and image compressing method based on this inpainting method is developed. These are of some certain practicability.