Boundary estimation in complex imagery using Fourier descriptors

A scheme is proposed to estimate the boundary in complex imagery by the use of Fourier descriptors. The rough boundary is encoded in vector form by connecting the centroid of the region and its boundary points and resampling angularly. Thus the 2-D boundary is transformed into a 1-D discrete curve. The discrete Fourier transform (DFT) is applied to the transformed and resampled 1-D curve, and the Fourier descriptors are used to estimate the boundary of interest. Only a few of the harmonics (low-frequency components) are used for reconstruction of the boundary, and by adjusting the number of harmonics, a good estimation is obtained. The shrink-expand operation is incorporated with the Fourier descriptors to reduce irregularities parts on the boundary and to improve the accuracy of the resulting boundaries.<>