Adaptive warped polynomial contour descriptors

This paper introduces a new set of discrete-contour descriptors, which are called improved Fourier descriptors (IFDs) and which are generalisations of the classical Fourier descriptors (FDs). A contour is represented by a complex-valued function which is approximated by a warped trigonometric polynomial (TP) of degree N. The IFDs are the 2N+1 coefficients of this TP. The paper shows theoretically and experimentally that the contour approximation using a given number of IFDs is more accurate than the corresponding approximation using the same number of FDs. In particular, IFDs represent visually important shape features much better than FDs. Therefore, they are promising shape descriptors in pattern recognition