Abstract
Image completion is a technique that removes an unnecessary part in the image, and restores the removed area (missing part) with keeping the consistency in appearance. Wexler et al. have formulated the image completion as an optimization problem. In the optimization problem, an energy function based on pattern similarities between the missing part and the rest of the image (data region) is minimized iteratively. The resultant image depends on varieties of patterns included in the data region. In this paper, we propose a new method considering rotated patterns of data region. The varieties of data patterns to be used in completion are augmented by rotating the image pattern of data region. Simultaneously, a local orientation of each rotated pattern is calculated. Thus, the suppression of computational cost can be achieved by the limited search utilizing the local orientation of rotated patterns. The effectiveness of the proposed method is verified by some image completion experiments.
