Auto-upgrading of signature case-base used for efficient detection of identity and authenticity

Handwritten signatures, used for identifying a person, suffer from two major drawbacks: they are inherently inconsistent, and they can easily be counterfeited. In this work a classifier is built applying Case-Based Reasoning techniques to resolve these issues by preserving authentic sets of off-line signature images of people within cases. For identification purpose, a bit-sequence containing the most frequent discretized mode value for each global feature is retained per person, and the corresponding pattern forms an index to the case pertaining to each person. Identity is established by finding the nearest matching case, while searching the base with discretized global feature pattern obtained from a test signature. Authentication of the test signature is achieved by comparison with weighted central metrics obtained from feature sets and dynamic time warping values. Essentially median vectors and inter quartile ranges are calculated and preserved for the cases apriori to serve as authenticity indices. As part of an incremental upgradation scheme, a newly authenticated test signature automatically gets preserved in the base, either as an additional specimen, or by replacing the worst of the existing lot if it qualifies as a better one, thus improving the discerning power of the system. The proficiency of the classifier is assessed through accuracy measurements on two sets of data-one downloaded from a standard database [19] and the other collected by the researchers.