Montran is constantly using Artificial Intelligence and other modern technologies to drive security-based and business decisions. In line with this legacy, our team is developing several applications using AI to simplify your automate business processes, improve security checks, make accurate forecasts, detect frauds and enhance customer experiences. In this blog, let us study the Montran proposition of an AI-based signature verification software to support banking and financial transactions.

SIGNATURE VERIFICATION
Signature Verification (SV) is the most popular method to ratify financial transactions, control access, and for a range of other offline banking services. As such, it is imperative that the signature verification system is able to identify the difference between genuine and forged signatures accurately. Physical signature verification is not only tedious but at times the human eye fails to detect differences that at times lead to fraud.

The test signature is usually on a cheque or a mandate or any document which needs pre-processing. Online signatures are acquired by means of a camera or a pressure sensing instrument while offline signatures are the scanned images of the signatures on paper. An ideal signature verification system must identify the forged signature by returning the similarity score to be very low as compared to a high score, obtained when two signatures perfectly match.

THE STORY SO FAR
There have been two approaches in the past for offline Signature Verification.

• Writer-dependent classifier – genuine signatures are labelled as positive and forged examples are labelled as negatives.
• Writer-independent classifier – finds the difference between the features of query signature against the genuine signature and the result is given.

Both of the systems have pros and cons depending on the use case.

A variety of offline methodologies and technologies have been used in the past such as:

• The offline signature verification system is built by a grid and tree-based feature extraction where features like permissible boundary, hand pressure, Euclidian distance are extracted.
• The language-specific signature verification system is designed in which Eigen-signature construction is deployed to form a feature vector.
• Artificial neural networks are deployed in which statistical features like Area, Centroid, Standard deviation, Even pixels, Kurtosis and Skewness are extracted.
• Fourier descriptors are used for online signature verification. The method is dependent on pen-up durations and drift and mean removal and the similarity score is evaluated by using Euclidean distance.
• Machine learning (ML) techniques are deployed wherein the statistical or geometrical, global and local features are extracted are fed to some ML-based models such as Support Vector Machines (SVM), Hidden Markov models or neural networks. It is observed that the performance of One-class SVM improves with the number of signatures.
• Local and global features are extracted in and are experimented on different systems. It is demonstrated that most of the local feature-based approaches take an enormous amount of time to compute the results.

It is found that these machine-learning based techniques require more than one genuine signature (either a natural or artificially generated) as well as forged or mismatched samples of signatures to train the system. Overfitting of the model with augmented data is also possible. This approach is not feasible for the BFSI sector.

THE MONTRAN TECHNOLOGY PROPOSITION
Montran proposes an alternate approach – using the pattern recognition technique on the pre-processed signature image. This image is matched with the one in the bank’s database and the similarity score is obtained. Our methodology also finds the forgery or mismatch of the signatures. Experimental results show that our methodology is more suitable for banks as machine learning based solutions require more than one signature per customer in the database or artificially generated signatures for matching.

THE WAY FOWARD
Signature verification system for the financial sector requires precise and swift decisions. It however faces the limitation on the use of online dynamic signature verification. With Montran’s accurate and quick signature verification solution, banks will be able to seamlessly manage massive volumes of transactions with efficiency and reliability.