A system that is able to infer color and texture information from a damaged two-dimensional (2D) or three-dimensional (3D) image is presented in this paper. The description in the paper goes beyond this, and shows a system that allows the user to choose any semantically representative object, remove it from a scene, and reconstruct the hole this object would leave.
A method called tensor voting is the core of the system. The authors provide results that are just superb. Only a few cases have not been properly inferred, namely, images in which small- and large-scale features lie within the hole. The method is stated to have an O(n) computational complexity, meaning the completion time is linear with the total number of pixels in the image.
The paper is very well structured and written, but it is hard to follow, due to the complexity of the methods involved. Only those experts with extensive knowledge of tensor voting will be able to completely comprehend the whole discussion. The paper is not recommended to any beginner in the field of reconstruction of images. The specificity of the paper should not be taken as a flaw; it is a major development in the field of image reconstruction and restoration.