Stefan Roth
Assistant Professor, Department of Computer Science, TU Darmstadt
Sample menu:
News:
Heinz Maier-Leibnitz-Prize 2012
PhD scholarship available
Masters progam in Visual Computing at TU Darmstadt
Inpainting Results
We evaluated the FoE model on two the image inpainting applications:Text removal
In this application we remove text that occludes parts of the image. The inpainting algorithm is given the location of all occluded pixels, which are found based on the color of the text.
Image with overlaid text:
Inpainting result from CVPR paper (PSNR 29.06dB):
Inpainting result from Bertalmío et al. (PSNR 27.56dB):
Masked image vs. inpainting result from CVPR paper:
Inpainting result from CVPR paper vs. inpainting result by Bertalmío et al.:
Scratch removal
In this application, scratches and other artifacts are removed from a photograph based on a user-supplied mask.
Original photograph:
User-supplied mask:
Inpainting result from CVPR paper:
Inpainting result from Bertalmío et al.:
Original vs. inpainting result from CVPR paper:
Inpainting result from CVPR paper vs. inpainting result by Bertalmío et al.:
