One of the central problems in pattern recognition and computer vision is the description and matching of shapes. A shape matching system for planar regions (two dimensional objects) often involves segmenting the boundary of the object and then matching the segments.
This dissertation presents a method for deriving a new feature which can be used in matching partial boundaries of planar regions. This global feature, called an Isthmus, can be efficiently computed from the Medial Axis Transformation (MAT) of the object.
A shape matching technique using the Isthmus feature is applied to the partial boundary matching problem of jigsaw puzzle fitting.
Three results of this investigation are presented. First, a shape matching technique based upon a global feature, called an Isthmus, was developed. The Isthmus feature may also be useful in a broader class of image processing problems such as: the narrowing of arteries in medical applications, geographic images, the obstacle avoidance problem in robot path planning, and any application in which one needs to find the point of narrowest necking of a planar region.
A second result of this investigation was an algorithm to compute the Euclidean Medial Axis Transformation. The algorithm has the following desirable features: (1) It uses Euclidean distances. Therefore, the skeleton produced is rotation invariant within quantization effects. (2) It produces a connected skeleton for each jigsaw puzzle piece. (3) The algorithm produces the Euclidean elevations (for deriving the skeleton) in O(n$sp2$) complexity, whereas the conventional method for computing the actual Euclidean elevations is O(n$sp3$).
These techniques and algorithms were implemented in a set of computer programs and applied to fitting together jigsaw puzzle pieces. The third result is an effective solution to the jigsaw puzzle problem where the pieces are constrained to interlock. An illustration of an actual assembly of a 24 piece jigsaw puzzle by these programs is given.