Noise reduction for tiny contours in image sequence

The contours of remote objects in the image sequence obtained by an optical sensor in space were often tiny and inclined to be stained by noise. It would affect the results of some contour based algorithms. In order to sovle the problem a noise reduction technique for the tiny contour in image sequence was presented in this paper. In the technique every contour of the sequence were described in frequency domain by the Fourier Descriptors (FD´s). A set of Interacting Multiple Model (IMM) estimators were used to filter the noise in every frequency point of the contours FD´s. The contour after noise reduction was retrieved by inverting the filtered contour FD´s. The experimental results show the technique was useful and effective.