Image inpainting with Markov chains
- PDF / 2,184,318 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 75 Downloads / 228 Views
ORIGINAL PAPER
Image inpainting with Markov chains Arpad Gellert1 · Remus Brad1 Received: 20 September 2019 / Revised: 15 January 2020 / Accepted: 17 March 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract This paper presents a novel context-based image inpainting method. The proposed technique is applying Markov chains to restore the colors of objects from images affected by some external factors (like scratches or wipes) or partially covered by other objects. Thus, damages or unwanted objects can be removed from an image by replacing each pixel from such an area, based on the surrounding unaffected context information. Therefore, the restoration process is applied from the exterior to the interior, by using for replacement colors occurring with the highest probability in similar contexts. Since we use context information, the proposed inpainting technique can successfully rebuild details in images. We have compared our method with other existing inpainting techniques, and the results were better on some test images or comparable on others. Keywords Image processing · Inpainting · Context information · Markov chains · Image restoration
1 Introduction Inpainting is a technique which is replacing missing or affected parts from digital images or damaged films (regularly small areas). Three main categories of inpainting algorithms can be distinguished in the literature: structural, textural and combinations of both. All these methods are using the information from the unaffected areas in order to reconstruct the affected or missing areas. The structural inpainting is using geometrical operations to re-establish missing pixel colors. Textural inpainting algorithms are building up stochastic models based on the information from unaffected image areas and are using the obtained models to reconstruct the affected or missing areas, thus being able to restore textures, too. The combination of structural inpainting and textural inpainting in a hybrid approach can exploit the advantage of both methods. In this work, we are proposing a new context-based inpainting technique, which is relying on Markov chains to replace affected or missing image areas. The area which must be reconstructed is defined by the user through points, which are connected afterward by lines. The reconstruction is started from the exterior of the affected area, with pixels
B 1
Arpad Gellert [email protected] Computer Science and Electrical Engineering Department, Lucian Blaga University of Sibiu, Emil Cioran Street, No. 4, 550025 Sibiu, Romania
whose context (consisting in the surrounding pixels) is at least partially in an unaffected area. The unaffected context part is searched within a limited surrounding window. The affected or missing pixel is replaced with the color having the highest probability to occur in similar contexts or context parts. In the next iterations of the reconstruction process, the restored pixel colors can be used to restore other pixel colors. By using context information, the proposed metho
Data Loading...
