Optimal segmentation of handwritten Chinese signatures using wavelet transforms

We propose a new segmenting method for handwritten Chinese signatures based on the wavelet transform for signature verification. There are some differences in identifying a handwritten signature and in recognizing a handwritten character because there are meaningful features hidden in writing habits when an individual is signing his or her signature. These features exhibit themselves in the pen-down, pen-up, and in the corner of a stroke. Therefore the segmentation for identifying a signature and for recognizing a character should be different even though the same characters are involved. We propose to segment an input signature curve at the inflection points, and we locate the inflection points by detecting the zero-crossing points of the wavelet transforms of the input signature. Experimental results show that this new segmenting method has better segmentation capability than other methods that are usually used. (C) 1996 Society of Photo-Optical Instrumentation Engineers.