Key Concepts

Before you learn about the use cases and implementation of Identity Scanner, it’s important to understand its fundamental concepts in detail.

E-KYC

The E-KYC or Facial Comparison feature of Identity Scanner compares the faces detected in two different images for matches in their facial features. The two images provided for comparison are termed as the source image and the query image. In the case of Identity Scanning, one would be the image detected in an individual’s ID proof, and the other would be a new or existing photograph of that person.

Facial Comparison Process

Zia AI implements various state-of-the-art object detection, image analysis, and pattern recognition models. A generic Facial Comparison process can be divided into the following sub-tasks:

• Face Detection: Face detection is a substantial part of the facial recognition process, which is infact a subset of object detection. This is the initial step towards facial recognition. The underlying algorithms detect faces in an image by segmenting them from the background, and identifying their coordinates in the image.

• Facial Landmark Detection: Facial landmarking is the process of detecting and localizing specific key point characteristics on the face to recognize faces. This includes identifying the coordinates of key points, such as the distance between the eyes, the length of the face, the endpoints of the face, the contour of the lips, ears, and chin, and the shape and position of the cheek bones.

• Facial Analysis: Facial comparison essentially involves comparing faces detected in query images to training sets or databases to verify an individual’s identity. In the case of Zia Facial Comparison, the faces detected in the query and the source images are analysed. The faceprints identified from the analysis data of both faces are compared with each other to verify the degree of matches. The confidence score that is returned by Zia corresponds to this data.

Various ML and deep-learning algorithms belonging to one or more face detection and face recognition methods can be used in performing these operations. These include geometric or feature based approaches, such as template matching, appearance or statistic based approaches that involve algorithms or methods like PCA using Eigenface or LDA, Fisherface algorithm, and implementation of neural networks.

Request Format

The input file format of E-KYC or Facial Comparison is specified below.

Response Format

The response format of E-KYC or Facial Comparison is described below:

• Using the SDKs and APIs:
  
  JSON response: The default JSON response will include the status message, and the data of the operation. Facial Comparison returns a binary response as to whether the faces match or not. The result is set to true if the faces match, or false if they don’t match. The JSON response contains a confidence score between the range of 0 to 1. Only if the comparison yields a confidence score of above 50% (i.e. 0.5) will the result be set to true. For more information, refer to the Java, Node.js and Python SDK, and the API documentation.
• In the console:
  
  When you upload images or files to test in the console, it will return the response in two formats:
  
  • Simplified response: This will display the message “Faces matched!” or “Faces not matched!” based on the results, and the confidence score of the comparison will be displayed in a percentage value.
  • JSON response: The console will also display the original JSON response received for the operation.

Document Processing

Text Recognition and Information Extraction Process

The text detection, recognition, and extraction from the identity proof documents is handled by an advanced OCR system. OCR systems in general follow a top-down approach to the process.

When a supported document is submitted to Identity Scanner, the text detection and recognition process proceeds as follows:

1. Zia analyzes the structure of the text and divides it into individual lines of text.
2. The lines of text are further broken down into words, and each word into individual characters.
3. Zia compares the characters it has detected with its dataset, and runs advanced algorithms to identify the characters and recognize words based on the character groupings.
4. Zia also runs algorithms to identify the language of the content by processing it through volumes of probabilities and hypotheses using ICR. You can additionally specify the language a document is in. This is only supported for the PASSBOOK model. When the language is already specified, this processing is performed more quickly. You can learn about the supported languages in the next section.
5. Through an extended OCR process, relevant identified information is extracted and populated as the values of pre-determined keys, based on the document model type. For example, values such as bank name and account number are recognized and extracted from bank passbooks and returned in the response with the corresponding keys.

Model Types

Identity Scanner supports the following model types corresponding to the supported identity proof documents:

• AADHAAR: You can process Indian Aadhaar cards as identity proof documents with this model type. You must provide two images of the Aadhaar card as the input (the front and back images of the card). The response will return the parameters recognized from the Aadhaar card, such as the Aadhaar number, the card owner’s name, address, and gender, along with confidence scores for each recognition.
• PAN: This enables you to process Indian PAN cards as identity proof documents by providing a single image of the PAN card as the input. The response will return parameters recognized from the PAN card, such as the card owner’s first name, last name, date of birth, and their PAN number.
• PASSBOOK: You can process the front page of Indian bank passbooks as identity or financial proof using this model, by uploading the passbook’s front page image. The response will return the parameters recognized from the passbook such as the account number, bank name, branch, and address. The response also shows if RTGS, NEFT, and IMPS have been enabled for that account.
• CHEQUE: This model enables you to process Indian cheque leaves as financial proofs, by uploading an image of the cheque. Identity Scanner only processes cheques of the CTS-2010 format. The response will return the parameters recognized like the amount, bank name, branch name, account number, IFSC code, and the date of issuance.

You must pass the respective model types in the request while using the Java, Node.js or Python SDK code , or when working with the APIs.

Supported Languages

PASSBOOK

The PASSBOOK model supports the following 9 international languages:

• English
• Arabic
• Chinese
• French
• Italian
• Japanese
• Portuguese
• Romanian
• Spanish

It also supports these 10 Indian languages:

• Hindi
• Bengali
• Marathi
• Telugu
• Tamil
• Gujarati
• Urdu
• Kannada
• Malayalam
• Sanskrit

AADHAAR

The AADHAAR model can detect and recognize textual content in all supported Indian languages and English. You can process the Aadhaar card of any region in India with Identity Scanner.

CHEQUE and PAN

The CHEQUE and PAN models process the content only in English by default, and do not support any other languages.

You must note the following points:

• You can pass the languages you might find in the bank passbooks while processing for the PASSBOOK model.
• Specifying the languages is optional. However, the text recognition and extraction process is faster and more accurate if you specify the languages. If no languages are specified, Identity Scanner will detect the languages automatically.
• You will not be able to select the languages for the CHEQUE or the PAN models, as English is the only supported language for these types. You will not be able to select the languages for the AADHAAR model either.

Request Format

The input file formats of each of the model types are specified below.

Response Format

The response format of Document Processing is described below:

• Using the SDKs and APIs:
  JSON response: The default JSON response will include the status message and the data for the respective operation, as specified in the Model Types section. You can format the text you receive in the response for the document processing in your code. For more information, refer to the Java, Node.js and Python SDK documentation.
• In the console:
  When you upload images or files to test in the console, it will return the response in two formats:
  
  • Simplified response: The simplified textual response will list the parameters extracted from the identity proof in a readable format.
  • JSON response: The console will also display the original JSON response received for each operation.