Offline signature Recognition System
Introduction
Biometrics can be classified into two broad categories-- behavioral and physiological. Handwritten signature is a behavioral biometric. It is amongst the first few biometrics to be used even before the advent of computers. Handwritten signature has long been used in the financial domain for identity verification. In most of the places the verification is done manually either by a person who is familiar to the signature or by matching it against a few signature templates (1-3 in number). Ramesh and Murthy claim that even a human verifier cannot be perfect. They claim 18% FRR, 0% FAR in case random forgery and 25% FAR in case of skilled forgery.
As mentioned before handwritten signature verification can be classified into offline signature recognition system (where only the static features are considered) and online signature recognition system (where on the contrary both the static & dynamic features are considered). Between the two, online signature recognition systems are more reliable because of its higher efficiency in terms of accuracy (closer to 99%) and time than offline (at most 90-95%). However, offline signature recognition systems cannot be ignored, since its applicability and ease of use are more in comparison to online signature recognition systems in many parts of the world ( e.g. on cheques, offline signature verification will be more useful than online). Moreover, online signature verification methods require some special hardware like digitizers, pressure sensitive tablets to capture the dynamic features, which the offline verification methods do not need to.
Methodology
The algorithm used for the implementation of offline signature verification systems consist of five major modules
Data Management: This module handles the various aspects of database management like creation, modification, deletion and training for a signature instance. The information regarding a particular signature is stored in the database as a feature vector.
Preprocessing and Noise Removal: Preprocessing in both offline and online generally involves removing noises like spurious pixels (in case of offline) or signals (in case of online), smoothening, space standardization and normalization, skeletonization, converting a gray scale image to a binary image, extraction of the high pressure region images, etc.
Feature Extraction and Parameter Calculations: Features can be classified into two types-- global and local features, where global features are characteristics, which identify or describe the signature as a whole (Ismail and Gad 2000) and local features are confined to a limited portion (e.g a grid) of the signature. Examples of global features include width and height of individual signature components, width to height ratio, total area of black pixels in the binary and high pressure region (HPR) images, horizontal and vertical projections of signature images, baseline, baseline shift, relative position of global baseline and centre of gravity with respect to width of the signature, number of cross and edge points, slant, run lengths etc. On the other hand, examples of local features are the same as that of the global features except that they are calculated for each of the number of grids that the signature has been divided to.
Learning: In the learning module we use the extracted features and calculate the mean and standard deviation for each of the feature. In case of some features such as the number of components in the signature and the slant of the signature mean and standard deviation do not have proper meaning. In this case the value of the majority feature is calculated e.g. in case of the number of components the median value is picked. These values are placed as a vector and stored in the database against the entered identification number. The current learning module is very flexible; so the number of training samples can range from 1 to the number which gives a stable mean and standard deviation. Although, a higher accuracy is obtained when we use more training samples.
Verification/Recognition: The recognition module compares the different features obtained from the image given to signature recognition system with the features stored in the database against the given identification number. Based on this comparison, it either accepts the signature instance as genuine or rejects it as a forgery.