AN ALGORITHM FOR THE IMAGE COMPLETION BASED ON THE EXAMPLES WITH A CHANGED CONFIDENCE FUNCTION

Abstract

Image completion (inpainting) is the process of recovering the missing or damaged parts of the image. The resulting image, after completion, should be visually plausible for the observer. The exemplar-based inpainting algorithm proposed by Criminisi et al is one of the most popular and effective algorithms in that sphere. This algorithm fills the unknown region by patches, searching for the most similar patches from the known region. A priority function is calculated for all boundary points, i.e. those points which are from the unknown region and have neighboring points from the known region. A fixed size patch, whose center is a point with the highest priority, is selected, and unknown points from that patch are filled. The priority function consists of two components: confidence term and data term. The confidence term depends on the number of the known points in the patch, as well as whether they were known from the start or were inpainted during the algorithm. The data term shows the structural characteristics in that point. The priority of patches to fill is critical in that algorithm, because due to its texture and structure, the information of the damaged image is recovered. The confidence term, decreases very fast during the update stage and this fact harms the result in some cases. In this paper, we suggest, at updating the values of the confidence term, to update them with a fixed value. As experimental results show, this approach gives better output in many cases. The proposed approach implementation is given and some comparison results of the algorithms are given.