A Novel Exemplar-Based Image Completion Scheme with Adaptive TV-Constraint

Image completion is an important task in the field of image processing with broad applications in image and vision analysis. we consider the image completion problem arising in image restoration: part of an image has been removed, and we want to restore the image from the remaining, possibly noisy, portion. In this paper, we present a new approach based on exemplar-based completion model which combines the characters of in painting and texture synthesis. Firstly we decide the size of image template based on the gradient domain of the image. Secondly, the filling priority of the template is determined by the curvature of the level lines and the confidence item of the template. This is also interpreted by total variation (TV) method of partial differential equation (PDE). Because of considering the geometrical property of the level lines, the structure property of the image is preserved. Thirdly because of the gradient domain denoting the property of the texture, when comparing the similarity between the source template and destination template, not only we compute the sum of squared differences (SSD) of the all ready filled pixels but also combining the information of the gradient domain of the templates. Lastly we copy the determined source template to the destination template and update the information of the destination template. Some experiments will show the advantage of our method.