• Image inpainting is the process of seamlessly filling in holes of arbitrary topology in an image to preserve its overall continuity. It is an ancient art of fixing accidental damage and recreating lost information.
• Object removal or modification in the original images can be carried out through image inpainting methods.
• In this project, various algorithms of Partial Derivative Equation based and Exemplar-based families have been studied and implemented. Results using Total Variation (TV) and Curvature Driven Diffusion (CDD) methods show that CDD produces a better visual quality of results. However, it fails to restore texture information.
• To solve this problem, Exemplar-based algorithms are studied and implemented. Traditionally, the data term present in this algorithm is based on the strength of the isophote found using the gradient. The problem with the gradient operator is studied, and a better contour preserving data term is proposed. The proposed data term uses the strength of structure line found using Infinite size Symmetric Exponential Filter (ISEF). This filter helps overcome the drawback of which overcomes the drawback of insensibility to noise and precision of edge localization present in traditional data term.
• Results are compared by quantitative analysis using PSNR, SSIM, and FSIM. Subjective analysis is done using Mean Opinion Score. It is proved that the proposed method produces better visual results compared to few other existing exemplar-based methods.
• Methods/Keywords: Exemplar-based Image Inpainting, PDE-based Image Inpainting, ISEF Filter, Priority Computation, Isophote, Curvature Driven Diffusion • Software/Tools/Programming Language Used: MATLAB, C
pratikgirigoswami/Exemplar-Based-Image-Inpainting
• Image inpainting is the process of seamlessly filling in holes of arbitrary topology in an image to preserve its overall continuity. It is an ancient art of fixing accidental damage and recreating lost information. • Object removal or modification in the original images can be carried out through image inpainting methods. • In this project, various algorithms of Partial Derivative Equation based and Exemplar-based families have been studied and implemented. Results using Total Variation (TV) and Curvature Driven Diffusion (CDD) methods show that CDD produces a better visual quality of results. However, it fails to restore texture information. • To solve this problem, Exemplar-based algorithms are studied and implemented. Traditionally, the data term present in this algorithm is based on the strength of the isophote found using the gradient. The problem with the gradient operator is studied, and a better contour preserving data term is proposed. The proposed data term uses the strength of structure line found using Infinite size Symmetric Exponential Filter (ISEF). This filter helps overcome the drawback of which overcomes the drawback of insensibility to noise and precision of edge localization present in traditional data term. • Results are compared by quantitative analysis using PSNR, SSIM, and FSIM. Subjective analysis is done using Mean Opinion Score. It is proved that the proposed method produces better visual results compared to few other existing exemplar-based methods. • Methods/Keywords: Exemplar-based Image Inpainting, PDE-based Image Inpainting, ISEF Filter, Priority Computation, Isophote, Curvature Driven Diffusion • Software/Tools/Programming Language Used: MATLAB, C
MATLAB