MULTISCALE B-SPLINE WAVELET FOR EDGE-DETECTION

Based on the theory of B-spline, a new Family of multiscale wavelet transforms has been presented. The edge of signals can be efficiently represented and detected through its zero-crossing or modulus maxima. For B-spline of order n, the fast algorithms for decomposition and reconstruction have been derived. Also the impulse and frequency responses of the corresponding decomposition and reconstruction filters are given explicitly. In terms of time/frequeny localization it has been proved that cubic B-spline is nearly optimal for most applications. The results have also laid a basis for further applications to stereo vision matching, denoising, etc.