Hole Filling for 3-D View Synthesis Using Sparse Representations

The challenge of inpainting holes of color information in images has been widely addressed throughout the image processing research community. One of the most popular methods from Criminisi et al. is exemplar-based and assigns filling priorities during the inpainting process. Another method based on Sparse Representations, which become more and more popular in the domain of signal processing, was proposed by Mairal et al. This work combines the idea of filling priorities and Sparse Representations. Furthermore it improves the results by inpainting over multiple scales and taking advantage of depth information. In addition, an exemplar-based method (MS-SSD-D) has been developed that uses multiple scales and depth as well. Whereas the results of the method using Sparse Representations (MS-KSVD-D) lie within the ones of existing algorithms, the MS-SSD-D method outperforms the available state-of-the-art algorithms in both objective and subjective quality measures of selected test images. The main application area of the developed inpainting algorithms are the type of holes, that occur during 3-D view synthesis (typically large vertical holes).     «

The challenge of inpainting holes of color information in images has been widely addressed throughout the image processing research community. One of the most popular methods from Criminisi et al. is exemplar-based and assigns filling priorities during the inpainting process. Another method based on Sparse Representations, which become more and more popular in the domain of signal processing, was proposed by Mairal et al. This work combines the idea of filling priorities and Sparse Representatio...     »