Bank Statement

This section describes how to build your custom OCR API to extract data from Bank Statements using the API Builder. A Bank Statement is an official summary of transactions that occurred within a given period.

Prerequisites

You’ll need at least 20 Bank Statement images or pdfs to train your OCR.

Define Your Bank Statement Use Case

Using the Bank statement below, we’re going to define the fields we want to extract from it.

• Full name: The full name of the client
• Address: The address of the client
• Account Number: The account number
• Opening Balance: The opening balance of the account at the beginning of the month
• Closing Balance: The closing balance of the account at the end of the month

That’s it for this example. Feel free to add any other relevant data that fits your requirement.

Deploy Your API

Once you have defined the list of fields you want to extract from your Bank Statement, head over to the platform and follow these steps:

1. Click on the Create a new API button on the right.

2. Next, fill in the basic information about the API you want to create as seen
   

3. Click on the Next button. The following page allows you to define and add your data model.

Define Your Model

There are two ways to add fields to your data model.

• Upload a JSON config File.
• Manually add data.

Upload a JSON Config

To add data fields using JSON config upload.

1. Copy the following JSON into a file.

{
  "problem_type": {
    "classificator": { "features": [], "features_name": [] },
    "selector": {
      "features": [
        {
          "cfg": { "filter": { "alpha": -1, "numeric": 0 } },
          "handwritten": false,
          "name": "full_name",
          "public_name": "Full name",
          "semantics": "word"
        },
        {
          "cfg": { "filter": { "alpha": -1, "numeric": -1 } },
          "handwritten": false,
          "name": "address",
          "public_name": "Address",
          "semantics": "word"
        },
        {
          "cfg": { "filter": { "alpha": 0, "numeric": -1 } },
          "handwritten": false,
          "name": "account_number",
          "public_name": "Account Number",
          "semantics": "word"
        },
        {
          "cfg": { "filter": { "is_integer": -1 } },
          "handwritten": false,
          "name": "opening_balance",
          "public_name": "Opening Balance",
          "semantics": "amount"
        },
        {
          "cfg": { "filter": { "is_integer": -1 } },
          "handwritten": false,
          "name": "closing_balance",
          "public_name": "Closing Balance",
          "semantics": "amount"
        }
      ],
      "features_name": [
        "full_name",
        "address",
        "account_number",
        "opening_balance",
        "closing_balance"
      ]
    }
  }
}

2. Click on Upload a json config.
3. The data model will be automatically filled.
4. Click on Create API at the bottom of the screen.

Manually Add Data

Using the interface, you can manually add each field for the data you are extracting. In our example, here are the different field configurations we used:

• Full name: type String that never contains numeric characters.
• Address: type String without specifications.
• Account Number: type String with no alpha characters.
• Opening Balance: type Number without specifications.
• Closing Balance: type Number without specifications.

Once you’re done setting up your data model, click the Create API button at the bottom of the screen.

Train Your Bank Statement OCR

You’re all set! Now it's time to train your Bank Statements deep learning model in the Training section of our API.

1. Upload one file at a time or a zip bundle of many files.
2. Click on the field input on the right, and the blue box on the left highlights all the corresponding field candidates in the document.
3. Next, click on the validate arrow for all the field inputs.
4. Once you have selected the proper box(es) for each of your fields as displayed on the right-hand side, click on the validate button located at the right-side bottom to send an annotation for the model you have created.
5. Repeat this process until you have trained 20 documents to create a trained model.

To get more information about the training phase, please refer to the Getting Started tutorial.

 

Questions?
  Join our Slack