Signature Verification Solution for FinTech Industry | Machine Learning | Case Study
azilen logo


Signature Verification Solution For BFSI Industry

About Client

Signature Verification is a task of determining whether a questioned signature matches known signature samples. Day in and day out, customers either drop cheques using drop box, or on cashier counter to pick up cash. These cheques are picked up periodically and submitted for clearing. This task involves checking if the cheque belongs to our bank, if yes, then get the account number, input the account number into the system and fetch customer details and signature sample. Next step involves manually inspecting the signature from the system against that found in the cheque. This procedure repeated over multiple times may cause burnout and can introduce manual error.

Business Requirement

  • Manual account number entry in system to fetch customer record.
  • Manual verification of signature in cheque against those stored in customer records.
  • Employee burn-out due to continuous execution of repetitive task.

The Solution

  • Signature verification is a technique used to validate the captured image signature to those signature image stored in records. It compares signatures and checks for authenticity. It generates a confidence score, where in a low confidence score suggest a forgery.
  • Machine learning refers to techniques used wherein intelligence is acquired by referring to examples. In order to solve the signature verification problem, learning Signature Verification Solution can be acquired in two ways:
    • Person-independent / general learning: it involves learning from a set of genuine and forgeries from a general population and differences between genuine and forged signatures across all individuals are learnt.
    • Person-dependent / special learning: it involves learning from multiple samples of only that person’s signatures where in within-person similarities are learnt.
  • It is observed that special learning performs better than general learning and there is an increase of up to 5% in the accuracy. Hence, we have developed our solution using special learning technique.
  • In order to perform automatic signature verification there are two parts :
    • Extract features from a signature
    • Determine similarities of two signatures based on these features
  • Based on the above methods, a confidence score is derived that indicates whether the question signature matched to that particular person or not.
  • Various scenarios like join accounts where more than one person is allowed to sign on the cheques, can be handled as well with custom logic.
  • In order to automate this process to an extent, a scanned copy of cheque will be obtained, machine will be able to recognize the account number and signature portion and carry on these steps seamlessly with existing systems.
  • Accuracy in the range of 80-85% was achieved.

Solution Benefits

  • Reduced time for cheque verification
  • Process automation
  • Lesser employee burnout doing repetitive task
  • Save time and energy and prevent human error
  • Prevent fraud