Robotics Report Technical Report

Excerpt from Robotics Report Technical Report

A technique for graphical assembly of large Jigsaw puzzles by computer vision is presented. The solution uses the puzzle piece shape information only. The pieces are photographed, digitized, their boundary information is extracted and a local matching algorithm is applied to compute the matching scores between every two sides of different puzzle pieces. Then a global matching algorithm, based on combinatorial Optimization techniques, is applied to compute the correct assembly of the puzzle pieces. The frame of the puzzle is solved by the traveling salesman algorithm and the the interior of the puzzle is solved by a branch and bound algorithm. The method was used successfully to assemble 104 piece puzzles, which have many pieces with almost similar boundaries. The technique was extended to assemble two intermixed 104 piece puzzles.