Shape recognition using segmentation of the boundary curve

Abstract

In this project a method is presented that enables a shift, ratation and scale independent representation of a 2-dimentional shape in the computer memory and a criterion is proposed for the comparison of two (or more) such shapes. The boundary curve is extracted and segmented in to such parts that can be represented each by itself with a single valued r(0) function. The FFI (Fast Fourier Transform) is of these curve segments are taken and the Fourier coefficients are truncated after certain value for the sake of data reduction. These truncated Fourier coefficients along with the number of curve segments of the shape and a set of values related with the angular span of each curve segment form a representation vector of the shape. The "distance" between such vectors forms a basis for the comparison of similarity between different sample shapes.