Our image completion algorithm automatically extracts mid-level constraints (perspective and regularity) and uses them to guide the filling of missing regions in a semantically meaningful way. Our method is capable of completing challenging scenes such as multiple building facades (left), strong perspective distortion (middle) and large regular repetitive structures (right). We significantly outperform three representative state-of-the-art image completion techniques for these images (see Figure 2). Input images (from left to right) are from © Flickr users micromegas, Theen Moy, and Nicu Buculei, used under the Creative Common license.
Figure 2: Limitations of current state-of-the-art methods. Compare these results with ours in Figure 1.
Sample comparisons with previous works
Visualization of the completion process in the coarse-to-fine fashion
Image credit: ©Flickr user Nicu Buculei
Image credit: ©Flickr users aigle_dore
Image credit: ©Flickr user cedrennes
Image credit: ©Flickr user the_dugghouse
Image credit: ©Flickr user sunshinepictures