Banking Industry Facial Recognition Data Collection

The Problem

A well-funded banking startup based in San Francisco, California, approached DataForce during the development of their facial recognition algorithm for secured payments and dual authentication methods. The algorithm specializes in the verification of biometric data, giving financial institutions added security detection and improving their ability to authenticate their customers’ identities through facial recognition data. In order to train the new algorithm, our client needed 2.2 million images and 400,000 videos from a globally distributed pool of races, nationalities, and ethnicities to acquire a diverse data set to train their machine learning model.

The Solution

DataForce began by collecting unique images and videos by skin type/ethnicity, size, duration, and type while also gathering 10 historical photos from each participant to train the algorithm with preexisting imagery of the individual. Using our global database of over 1.3 million contributors, DataForce collected the millions of photos and videos needed to train the algorithm, ensuring the submissions had varying gender, race, age, country of residence, and country of origin to build the diverse data set.

Aside from the diversity components, there were over 10 variations of lighting, poses, and props to decrease bias in the data. Examples of the variations included wearing dark sunglasses, baseball hats, cloth masks, and sweatshirts with the hood up and down in different lighting and paired with multiple camera angles. The millions of photos and videos were all attained and centralized through our internal application, DataForce Contribute, which allowed our client to collect the data without having to be present with the participant—all with safety and security of personal data at the forefront. Once DataForce successfully executed the data collection, the client was then able to add the diverse data sets into their models, allowing them to provide a more accurate and secure facial recognition model to their clients.