Image completion based on segmentation using neutrosophic sets

Abstract

Image completion aims to restore the corrupted regions in an image. One of the most important challenges in image completion is to find the most appropriate data for replacing in the hole. In this paper, we proposed a novel image completion method which applies neutrosophic-based segmentation to fill in the hole. Since neutrosophic is useful to interpret indeterminacy included in images, we applied neutrosophic-based image segmentation to decrease spatial and intensity ambiguities exist in images while we have more boundary and homogeneity preservation and less discontinuity. Our exemplar-based image completion algorithm starts from the outer pixel with the maximum priority and iterates until there is not any pixel in target region. Our extended similarity measure considers both neighbourhood and similarity based on results of our innovative neutrosophic-based image segmentation algorithm to find the patches with the maximum match for hole completion. Results show that our approach introduces an improvement of 18% for ASVS (Average Squared Visual Salience) compared to earlier methods. We also gained 0.9919 and 38.96 for MSSIM (Mean of Structure Similarity) and PSNR (Peak Signal to Noise Ratio), respectively while the best values for earlier methods were 0.9868 and 36.75 for MSSIM and PSNR, respectively which is the effect of using neutrosophic segmentation.