The Development of a Fragment-Based Image Completion Plug-in for the GIMP

Recent developments in the field of image manipulation and restoration have seen the merging of techniques that reconstruct texture and those that reconstruct structure to create algorithms that are effective for a far greater variety of image completion scenarios than were previously possible. Few of these innovative and useful algorithms are available to ordinary users however. We describe the implementation of a version of one such technique, the fragment-based image completion algorithm developed by Drori, Cohen-Or and Yeshuran [2003]. This automatic image completion algorithm uses the known parts of an image to infer the unknown parts, thus reconstructing the missing region as realistically as possible. We implement this algorithm as a plug-in for the GIMP making it freely accessible and easy to use in conjunction with other image manipulation tools. Furthermore, the process of developing plug-ins for the GIMP is evaluated and an on-line tutorial for writing general plug-ins for the GIMP is developed alongside the image completion plug-in implementation. Results achieved by this plug-in show that it is highly effective for completing regions that have random texture and it reconstructs the general structure of objects well. Geometric shapes are not necessarily completed as expected however because the algorithm does not take mathematical principles of shapes into account. It is concluded that the image completion algorithm is suitable for most of the scenarios put forward by the original authors. The tutorial provides a starting point for writing general plug-ins for the GIMP, covers common problems and lists useful resources. The process of plug-in development for the GIMP is well documented and supported.