PERFORMANCE ANALYSIS OF IMAGE INPAINTING WITH JOINT FILTERING OF MULTIPLE PATCHES USING SVD AND ALPHA TRIMMED FILTER

Abstract

Image inpainting is a smart editing tool. The patch-based algorithms inpainted the damaged/missing region patch by patch propagating from the boundary towards the center of the missing region. Several patch-based image inpainting methods relied on single patch selection and few on multiple patch selection-based method. Multiple source patches were used to obtain a more similar source patch for the target patch. Since the single source patch was chosen based on partially known target patch, a large variation resulted in artifacts in the inpainted region. To overcome it, multiple source patches were selected and filtered jointly to capture the core pattern amongst them. The resultant filtered patch was an optimal patch for the corresponding target patch. This paper compared the performance of joint filtering of SVD values of similar patches(KNN-SVD)and filtering of similar patches using alpha trimmed filters(KNN-KV-Alpha).These two algorithms were tested and evaluated for input images and results were compared with standard algorithms. Results showed that the time required for inpainting was drastically reduced while the quality factor was equivalent with the existing techniques. KNN-KV-Alpha gave high quality factor with less time for inpainting as compared to KNN-SVD as the similar patches were selected in the neighbourhood of the missing region