Bank Form Automation
Bank Form Automation
Bank Form Automation is powered by Deep Learning techniques to capture the key information available in various types of forms used at Banking Institutions. Such forms include
- Account Opening Forms for various products such as Savings account, current account, loans, credit cards, etc.
- Standing Instructions/NACH forms
- Standardized Order forms
Information, both printed and hand-written, extracted from these Forms include
- Customer Name
- Customer Address
- Customer Email
- Customer Email
- Phone Number
- PAN Number
- Aadhaar Number
- Citizenship data
- Gender
- Income details
- Education details
- Signature(s)
Why Bank Form Automation?
- Helps in extracting printed and handwritten data from forms so that the scanned document can be automatically transcribed.
- Reduces the error prone, time consuming manual effort to capture information from an Invoice.
- No need for defining the ROI when new formats are encountered, the product is built to automatically understand and capture information with a technology that mimics the Human Brain.
- Simple and Easy deployment with SDKs and APIs.
Most Salient Features are
- Save Costs
- Reduce Errors
- Improve Efficiency
- Scale at Will
How does Bank Form Automation work?
Bank Form Automation software captures information using advanced technologies from the field of Artificial Intelligence
1. Process Outline
The software takes a scanned image as input and performs custom built preprocessing steps to clear noise in image and make the sample suitable for information extraction. Then the system identifies the key-value pairs from the images. Keys are labels of information being captured. Values are data points that are filled in (either printed or written) by customer and could be any of the following
- Data written within a jail box
- Data written within a box
- Data written on a solid/dotted line
- Data written in a free form manner
- Selections within a range of checkboxes.
Below is a high level processing flow
- Keys across the documents will be identified
- ROI algorithms will identify Value fields
- A mapping algorithm will associate keys and values
- A preprocessing algorithm scans each value field to identify the type of the field (jailbox, box, line etc)
- Specific denoising algorithms are run on the value fields based on the field type
- Once the value fields are denoised, they are fed to the recognition engine to extract data
- Data extracted is then passed to appropriate post processing algorithms to validate data
At the end of the processing flow, all keys and corresponding values are extracted and assembled into a desirable format
2. Output Format
The information is captured as key value pairs for easy integration with a customer’s business process. The extracted information can be customized to customer requirements such as CSV, XML or JSON format.
3. Learning Ability
Bank Form Automation works much like a human reading an invoice, it uses cognitive skills to detect patterns, structures and candidate regions for information capturing. It defines the ROI based on the features it has learnt during past experiences.
4. Confidence against the captured information
The detection is further strengthened by giving a confidence score against each pair of information captured. The confidence score is a measure of how much the captured value is associated with the given key name.
5. Manual Intervention
Bank Form Automation performs correction and validation of data captured, using custom built algorithms. If there is any ambiguity in validation of information, Bank Form Automation automatically prompts the user to validate the information captured and review the data, add or change. This feedback from Human intervention helps Bank Form Automation to learn and become more accurate terms of information extraction.
How do we achieve near-100% accuracy?
Multiple algorithms for region-of-interest detection
In Bank Form Automation, there are multiple algorithms working simultaneously on region-of-interest identification, ensuring 100% detection.
Adaptive field-of-view
Before extraction, EazyForm accommodates inconsistencies in data placement, such as the date being slightly outside its boxes.
De-noising
De-noising systems from simple Otsu methods to deep learning-based segmentation algorithms improve Bank Form Automation’s extraction quality
Dual algorithmic journeys for building quorum
It builds quorum using two algorithms for every field. Each algorithm has different deep learning bases and maths for feature extraction methods, number of layers, loss functions etc.
Ensemble of algorithms for signature verification
Bank Form Automation’s algorithms are tuned to account for challenges of low image quality while verifying signatures with the KYC document accompanying the application form