A New Image Inpainting Approach based on Criminisi Algorithm

Abstract

In patch-based inpainting methods, the order of filling the areas to be restored is very important. This filling order is defined by a priority function that integrates two parameters: confidence term and data term. The priority, as initially defined, is negatively affected by the mutual influence of confidence and data terms. In addition, the rapid decrease to zero of confidence term leads the numerical instability of algorithms. Finally, the data term depends only on the central pixel of the patch, without taking into account the influence of neighboring pixels. Our aim in this paper is to propose an algorithm to solve the problems mentioned above. This algorithm is based on a new definition of the priority function, a calculation of the average data term obtained from the elementary data terms in a patch and an update of the confidence term slowing its decrease and avoiding convergence to zero. We evaluated our method by comparing it with algorithms in the literature. The results show that our method provides better results both visually and in terms of the Peak Signal-to-Noise Ratio (PSNR) and Structural SIMilarity index (SSIM).

Specular Reflections Detection and Removal for Endoscopic Images Based on Brightness Classification

“…This function will cause two problems in the actual repair: (1) The confidence item drops sharply and tends to 0 quickly, resulting in the random selection of repair sample blocks in the repair process. To avoid this phenomenon, Jing [ 29 ] introduced a regularization factor to control the smoothness of the confidence curve in the confidence item, and Ouattara [ 30 ] changed the multiplicative definition of the priority function to a weighted summation. (2) When the confidence term is large, the data term may be zero.…”

“…the highlights can no longer be considered sparse parts, and the method does not work well. Image inpainting-based methods [ 6 , 7 , 8 , 19 , 27 , 28 , 29 , 30 , 31 ] first detect the highlighted regions in the image and then inpaint with the most similar parts in the image. It can restore the specular regions under certain conditions, but it may be unreasonable for the reconstruction of human tissue structure, or the recovery effect for large-area specular regions is not good, or the algorithm operation efficiency is low.…”

Inpainting in discrete Sobolev spaces: structural information for uncertainty reduction

“…In the analysis phase, the scanning order to inpaint Ω seems to play a central role for the correctness of the reconstruction. In this direction, different works have tried to improve the confidence term firstly introduced in [12], to avoid the too fast convergence to zero of the original formulation (see [11,44,56] ). We shortly recall that in [12], the scanning order (also called priority) P (i, j) has been defined as…”

“…The contributions of C(i, j) and D(i, j) have been usually considered separated, as respectively explanatory of the uncertainty of the given data and the preferential direction of propagation of the isophotes ( [11,44]). Both C(i, j) and D(i, j) belong to the interval [0, 1], given a suitable normalization coefficient α (e.g., 255 if W=8).…”

An improved Criminisi method for image inpainting

“…It is widely used in fields like protection of historical relics [1], restoration of old photographs [1], biomedicine [2], etc. Methods for image inpainting can be roughly divided into diffusion-based methods and exemplar-based methods [3]. The diffusion-based methods use the pixels of the known region neighboring the region to be repaired to determine the structure and content of the diffusion.…”