Research Of Optimal Method Of Image Inpainting Based On Texture Synthesis

In a good many fields of computer graphics and computer vision, image inpainting is a very important research project and has been studied by a large number of scholars, and has a wide range of applications, such as the protection of antiquities, the repair of old photos, the removal of texts and obstacles in image, and the production of film and television special effects and so on. With the deepening of the study and the continuous expansion of the scope of application, this technology has more and more extensive attention from researchers, and various image inpainting algorithms have also emerged.This paper analyzes Criminisi image inpainting algorithm, and finds that its repair time is too much overhead and priority calculation exists a certain lack, moreover in the algorithm the module size is always a fixed value, so puts forward two optimal methods of different emphases, the main research work of this paper has the following two points:(1) A method is proposed, which ensures the image inpainting effect is not poorer than Criminisi algorithm, while improves the repair efficiency as much as possible. First, it calculates average gray values of all texture blocks in the known area of image to be inpainted in advance before SSD(sum of squared differences) exact matching, and in the matching process, several texture blocks in the known area having relatively large gray differences are eliminated by comparing their average gray values with that of the current block to be inpainted with threshold control, in order to save much matching time and accelerate the inpaint speed. Then, it defines one kind of new priority computation formula, which adds the influence from the factor whether close to original boundary on the priority, thereby that makes the priority calculation more reasonable and the result after inpainting better. Finally, the experiments prove that the image inpainting efficiency is increased greatly relatively, and the results after inpainting look like more natural and more fitful to people’s visual perception.(2)In the algorithm the module size affects the image inpainting effect to a great extent, therefore exploring choosing different module sizes for damaged images with different texture complexities is particularly important. Considering that the choice of the module size is related to the degree of complexity of the texture, so an image inpainting algorithm combining gray-level co-occurrence matrix and entropy is provided. It first makes use of gray-level co-occurrence matrix to extract image’s texture feature values, then completes the choice of module size according to the value of entropy. In addition, in order to solve the problem when there are multiple best matching blocks, Criminisi algorithm may be due to the improper choice while causes ineffective repair, when finding the best matching block, not only considering the differences of color information, but also taking into account the factor of spatial distance. Finally, PSNR measurement to evaluate image inpainting quality objectively is given. The experiments show that, compared with Criminisi algorithm, this method can choose appropriate module size neatly according to whether the texture of image to be inpainted is complex or not, and the inpainting effects are more natural and more fitful to people’s visual perception.