A New Exemplar-Based Image Inpainting Algorithm Using Image Structure Tensors

Digital image inpainting will be a standout amongst the practically imperative fields over image processing. Recently, exemplar-based inpainting strategies have drawn additional attentions due to their simplicity as well as their adequate computational cost. One of the most well-known and fundamental algorithms in this field has been proposed by Criminisi. In spite of the fact that the Criminisi’s algorithm is fast, simple and makes effort to propagate both image textures and structures, it fails when it comes to inpainting images with convoluted structures. To tackle this issue, in this paper two improvements to the Criminisi’s algorithm have been suggested. Image structure tensors have been employed to help image gradients out with image structures detection. This way, those essential structures are being distinguished from the ones with less degree of significance. Furthermore, the priority function in the Criminisi’s algorithm has been reformulated, so that the impact of the image structures is strengthened. Experimental results of the proposed method demonstrate its capability to overcome the stated problems. Moreover, inpainting quality has been significantly improved.