Faster Fragment-Based Image Completion

The fragment-based image reconstruction algorithm presented at SIGGRAPH 2003 by Drori et al. provides a way to complete missing parts of an image by iteratively approximating the unknown regions and compositing image fragments into the original image. We modified this algorithm in several ways, including the addition of a KD-Tree to speed up the fragment search process, and implemented the algorithm in Java using the Java Advanced Imaging APIs. We also used images of significantly higher resolution than the images used in the paper by Drori et al., therefore providing a challenging test of the algorithm’s effectiveness. Our experiments demonstrate that this modified algorithm does an effective job of reconstructing higher resolution images and improves upon the speed of the original. Empirically, we find that our implementation with a KD-Tree can perform at least 16 times faster than one without it.