A deformable wavelet descriptor and its application to curve extraction from noisy images

We developed a hierarchical planar curve descriptor which decomposes a curve into components of different scales by using the wavelet transform in our previous work. We interpret the wavelet coefficients as random variables and call it the deformable wavelet descriptor. The application of the deformable wavelet descriptor to boundary curve extraction from noisy images is examined. Due to the multiscale and local support properties of the wavelet basis, the deformable wavelet approach performs better than the traditional Fourier approach. Experimental results are given to demonstrate the performance of the proposed method in extracting the boundary curve of an object from low SNR images