Partial matching of two dimensional shapes using random coding

Abstract

Three algorithms for coding the boundary of two-dimensional shapes are described. In each algorithm, each boundary point is coded with respect to another point picked at random from the boundary. Using this method, an effective and efficient correlation technique to match two-dimensional shapes is developed. This technique can be used to match shapes of arbitrary scale and orientation. The given shape can have a closed or open boundary or even have a portion obstructed from the view. Matching can be performed with varying degrees of detail, giving the technique an added robustness, against geometrical distortions. It can also discriminate between different shapes. On an IBM 3033 computer it typically takes 10 CPU-seconds to generate one correlation curve between two shapes, each with a 500-point boundary curve.<>