Jigsaw Puzzle Solver to Locate Piece Position

Jigsaw puzzles present an interesting form of problem solving based around techniques of matching and feature recognition. This paper presents a method to find the location of jigsaw pieces relative to the completed puzzle using an ORB feature detector alongside a FLANN based matcher. An image or frame containing a jigsaw piece is reduced to its region of interest through thresholding and contour finding before being passed through the feature matching process. Matches are then filtered using Lowe’s ratio test and outliers are found and removed using a homography transformation matrix. Evaluation of the proposed method saw 81.2% accuracy of a jigsaw puzzle with 24 pieces which was significantly more accurate than shape detecting methods and more robust than methods which required specialised jigsaw pieces.