Solving Jigsaw Puzzles Using Computer Vision and Curve Similarity Measures

Jigsaw puzzles are a popular form of entertainment. Solving them with the help of computers raises several interesting problems, and has been the subject of many published papers in the past. In this paper, an approach to creating a program that uses pictures of real puzzle pieces to reconstruct the full puzzle image is presented. A photography technique is described that results in consistently recognizable images, which are used to extract features of the puzzle pieces. These features are then compared using two similarity algorithms - Hausdorff distance and Dynamic Time Warping. Three different puzzle assembly strategies that use these comparisons are presented, along with additional logic rules to solve the full puzzle. Using these the final program is capable of solving two different, 25-piece jigsaw puzzles. A comparison of the different similarity measures and assembly algorithms in the scope of the problem is also presented.