Veryfi API Fraud Detection Framework
Field Name | Comments |
Handwriting Manipulation
| Under the hood Handwriting Detection is a micro service in the data extraction process that looks at the selected fields and its areas and tries to identify whether there are any traces of handwriting. Veryfi Machine learning algorithms are trained on a labeled dataset of handwritten samples. The algorithms learned to recognize different handwriting styles and characters based on the extracted features.
Example: User has added a new line item by hand and/or modified the total by adding a 0 to the total to exponentially increase their spend claim. Many use cases would want to flag such receipt (main application is expense management and cpg loyalty campaigns)
|
Data Reconciliation
| Field called warnings is part of the meta object. The idea is that some of those signals can help our clients to detect fraud. Example :
|
LCD Detection
| Detects Moiré pattern due to interlaced scanning.
|
Detect Handwriting
Veryfi handwriting detection [handwritten_fields ] adds a layer of evidence to assess the authenticity, integrity, and intent behind a document. Handwriting detection plays a significant role in forensic examinations and fraud investigations. Handwriting analysis can detect alterations or additions made to documents. Fraudsters may modify numerical values, names, or other critical information in an attempt to deceive others.
Under the hood handwriting detection is a micro service in the data extraction process that looks at the predefined fields and their areas and tries to identify whether there are any traces of handwriting. Veryfi Machine learning algorithms are trained on a labeled dataset of handwritten samples. The algorithms learned to recognize different handwriting styles and characters based on the extracted features.
Example: User has added a new line item by hand and/or modified the total by adding a 0 to the total to exponentially increase their spend claim.
What data handwriting detection looks at
Out of the box handwritten_fields look at the particular number of fields from a document [total, subtotal, date, line_items].
The most up to date information is available in API schema inside Interactive API
JSON Sample
"meta": {
"handwritten_fields": [
"date",
"line_items.0.total",
"total"
}
Computer Screen Detection
is_lcd field - LCD (Liquid Crystal Display) detection in images refers to the process of identifying whether an image is a screenshot of an LCD display. This task is performed to detect forged or manipulated images, especially in scenarios where screenshots are used to deceive or misrepresent information.
LCD detection is beneficial in detecting fraud, particularly in cases where screenshots of an LCD display is being used to forge or manipulate invoice or receipt data.
Veryfi detection algorithms can analyze the spatial, temporal, texture, color, or light reflection characteristics of an image to identify any inconsistencies or artifacts that indicate manipulation. This helps detect forged or altered invoice or receipt images.
Use case: Fraudsters may modify key details on invoices or receipt, such as the invoice amount, vendor information, or payment terms, bank account details, to deceive recipients into making incorrect payments or to inflate expenses. By altering the data on an invoice, fraudsters can claim reimbursement for expenses that were never incurred or seek refunds for non-existent purchases.
Incorporating Veryfi LCD detection techniques into a broader fraud detection framework strengthens the overall capability to detect and prevent fraudulent activities related to invoices and receipts. By combining LCD detection with other fraud detection methods, like Detect Handwriting, Data Reconciliation, Detect Blurry, Duplicate Detection and Document Classifier users can improve their ability to identify and flag suspicious submissions.
Here are a few other scenarios where LCD detection can aid in fraud detection:
Document forgery: Fraudsters may create fake documents or screenshots to mislead others. By applying LCD detection algorithms, suspicious screenshots can be analyzed for inconsistencies in texture, color reproduction, or light reflection, which can indicate that the document has been manipulated or forged.
Financial fraud: Screenshots of banking transactions, payment receipts, or investment statements can be used to deceive individuals or organizations. LCD detection can help verify the authenticity of such screenshots, identifying any signs of tampering or fabrication.
There are various techniques and algorithms used for LCD detection in images. Here are a few common approaches:
Spatiotemporal Analysis: LCD screens emit a unique pattern of flickering or temporal changes that can be captured in a screenshot. By analyzing the spatial and temporal characteristics of an image, it is possible to detect these patterns and determine if the image originated from an LCD display.
Texture Analysis: LCD screens have a distinct texture caused by the arrangement of pixels. This texture can be extracted from an image and compared against known LCD patterns. Deviations from the expected texture can indicate that the image is not a genuine screenshot.
Light Reflection Analysis: LCD screens produce specific patterns of light reflection due to their polarizing filters and backlighting. By analyzing the distribution and intensity of light reflections in an image, it is possible to determine if the image was taken from an LCD display or another source.
Color Analysis: LCD screens have unique color reproduction characteristics, such as color gamut and color temperature. Analyzing the color properties of an image, such as color consistency and color distortion, can help identify if it originated from an LCD display.
JSON Sample
"meta": {
"fraud": {
"images": [
{
"is_lcd": true,
"score": 0.86
}
}
After December 1, 2023 field name changes to meta.fraud.pages
"meta": {
"fraud": {
"pages": [
{
"is_lcd": true,
"score": 0.86
}
}
Data Reconciliation
A field called warnings is part of Veryfi fraud framework. The idea is that some of those signals can help our clients to detect fraud. warnings are part of the Reconciliation and Integrity check Veryfi provides. Integrity checks are used to verify the integrity and authenticity of invoices and detect potential fraudulent activities. These checks help ensure that invoices have not been tampered with or manipulated.
Integrity checks are often performed manually by trained professionals or can be automated through software systems that compare and analyze invoice data against predefined rules and patterns. By conducting these checks, organizations can identify potentially fraudulent invoices, prevent financial losses, and maintain the integrity of their financial processes.
Response Sample for warnings
"meta": { "warnings": [
{
"message": "Line item Description \"Cocacola\" repeats on lines 2, 3",
"type": "line_item_repeats"
}
List of all possible types of warnings
"tax_rate_missmatch" | when Tax Rate x Subtotal != Tax Amount |
"item_counts_missmatch" | when number of products on line items doesn't match Number of Items Sold value |
"totals_missmatch" | when subtotal from line items doesn't match subtotal on the document |
"line_item_amount_missmatch" | when quantity x price != line item total |
"line_item_repeats" | when items repeat (SKUs or descriptions) on document multiple times |
"barcode_decoding_issue" | when barcode is present on the document but is not decodable |
"logo_vendor_missmatch" | when vendor name from logo doesn't match vendor name output from model |
"barcode_code_missing_in_ocr" | Decoded Barcode numbers were not found on the document |
Most recent info can be found in API schema inside the interactive APIs