Signature pattern recognition using pseudo Zernike moments and a fuzzy logic classifier

We have introduced a new method, taking advantage of an image moment transformation combined with a fuzzy logic approach. For this purpose first we tried to model the noise embedded in signature patterns inherently and separate that from environmental effects. On the basis of the first step results, we have extracted the most optimum mapping to a unit circle using LMS criteria. Then we derived some orientation invariant moments introduced in former reports and studied their own statistical properties in our special input space, using a new defined criterion. Afterwards we defined an error matrix for signature patterns and studied its behavior and concluded that a fuzzy classifier seems to be the best choice for our application. Then we defined a fuzzy complex space and also a fuzzy complex similarity measure in this space, and constructed a training algorithm to learn the fuzzy classifier. Thus any input pattern could be compared to the learned prototypes through a pre-defined fuzzy similarity measure and attributed to one of the learned classes. The fuzzy classifier is applied to each of the above derived moments which constituted an individual feature space separately and miss-classifications were detected as a measure of the error magnitude. Finally a comparison is made between the above considered image transformations and we have pointed out some of the advantages of this method.