It is still too easy to trick people with pictures... We miss simple, smart, efficient plug-ins to support our processes. (CFO)

Use Cases & Pain Points Addressed

This tool solves the following pain points, or greatly reduces their impact:

In principle, fraud cases depend on the business needs/use cases of the customer. This platform could be used in an endless number of scenarios, all of which...
  • Require a comparison of a given image published in digital media, or submitted in a request, with an available pool of ""own pictures"" to find matches.
  • Require extraction of text from pictures in digital media, or submitted ones, and their subsequent intelligent interpretation.

Implemented use cases:
  • Help insurances in uncovering fraud by analysing the pictures of accidents, and matching them vs. a database of other accident pictures, in order to find recurring claims filed with the same picture.
  • Help banks in uncovering fraud in the financing second-hand cars: loans are being given according to the value of those proposed cars, but which are wrecked in real. The money is instead being used for other purposes.
  • Crawl through the news of accidents in digital media (national and regional):
    - Matches license plates of cars involved in the accidents in order to identify possible fraud cases. This allows for tracing the real cause of the accident.
    - Parses articles to give clues on whether the driver was (e.g.) under alcohol influence, even though the accident was filed in another context.

More examples:
  • E-commerce sites may detect theft in product postings.
  • Real-Estate companies may detect sales objects copied by competitors to attract attention from potential buyers, then negotiate a share of the sale.
  • ... Any company whose business critically relies on picture support.

    Key Features & Differentiators

    The following features and advantages should be noted:

    • Features a specially developed technique/algorithm that allows for very high scale and speed.
    • Some other image matching programs exist, bur not close to the processing speed and cannot scale that much (i.e. limited to matching thousands of pictures, but not millions).
    • Editing features that can alter a picture via Photoshop are being detected as well (except to some extent: crop), therefore attempts to trick attention through fraudulent editing are being uncovered.
    • Additional web crawling and digital processing of online information makes use of machine learning techniques for intelligent and enhanced processing of news data.

    ... In principle, the specific fraud scenarios of the customer determine the amount of customization and training required by the tool, i.e. to adapt and integrate with its business process. This effort should be considered as being a small project.

    IT cost and footprint is very low: a simple i3 laptop can process 50 Mio. comparisons in 1 minute. (e.g. 10 pictures can be matched with a pool of 5 Mio. pictures, in 1 minute).

    KPI examples from implemented use cases:
    • 50% of the alcohol-related accidents are found to have been claimed fraudulently, and are being detected within 1 week of the accident.
    • In one medium-sized bank, having a market share of just 1%: 8 cases were found in a single month, related to fraudulent loan requests for second hand cars.


    Cannot be disclosed.
    • 8am - 4pm
    • 10am - 2pm