Study Of Digital Image Inpainting Algorithms

With the advent of the digital age, many precious photos were scanned and saved on the computer. However, these photos always have some defects due to some natural or unnatural factors when preserving. Besides that, sometimes we need to selectively remove some parts of the picture which are not interested in. Under this condition, the digital image processing technology emergences and gets more and more attention.Digital image inpainting is to repair the damaged areas conveniently under the guidance of some principles and algorithms, it needs to be finished from the perspective of human vision. Its target is to make the results more clear and natural, and make the viewer cannot aware of any modification.Digital image restoration algorithm can be divided into two general directions, the repairing techniques for recovering small-scale damages, such as the scratches and spots, and the completion technology for filling the large-scale blank region after removing some objects. Currently, the technology for small-scale restoration is relatively mature, so there is no specific description in this thesis. This thesis mainly focused on how to repair a large-scale region, and discussed the principles, processes, and effects of the following algorithms.Firstly, we introduced the background and significance of image inpainting technology, and summed up the research status of different inpainting techniques. Then we in-deeply discussed the classic texture synthesis algorithm, and on this basis proposed a non-regular texture image completion algorithm in large region. This algorithm improved the restoration order by adding a directional priority coefficient, speeded up the efficiency and reduced the mismatch by limiting the searching scope of best-matching patch.Besides that, we proposed a structure-based and the decomposition-based algorithm for the restoration of images with obvious structure information. Since the human eyes are more sensitive to the structural information, the structure was connected and restored firstly, and then the connected structure was utilized as the watershed for the image partition. Finally the remaining texture information in different parts were filled separately. Experimental results show that the algorithm can achieve good results for complex structure images.In this thesis, we proposed a decomposition-based restoration algorithm by introducing the technique in the image inpainting field. By decomposing an image into two parts of texture and structure, we can inpaint them respectively. More over, the decomposition technique can also be used to improve the structure propagation algorithm. Finally, we analyzed the advantages and disadvantages of the algorithm with the experimental results in this thesis.