CAUSES OF LOW SUCCESS RATE: VERITAS FACIAL RECOGNITION PROCESS FOR REVOLUT Face detection algorithms can detect a human face from inside a detailed scene. A face on a person is about as unique as you can get when you want to identify a person based on one outstanding feature. Facial recognition works by identifying a person from a digital image or video source by comparing and analysing patterns with particular regard to the contours on the person’s face. It provides safe and highly reliable security mechanisms especially for corporates and governments. Compared to technologies and biometric systems like palm prints and retina scans face recognition has been proven to be more effective as a silent mechanism which doesn’t need human interaction for it to work. It can be done without a person’s knowledge thus the capability of security offices to keep track of human activities in an airport or and the face_recognition library: # import the libraries import osimport face_recognition # make a list of all the available images images = os.listdir('images') # load your image image_to_be_matched = face_recognition.load_image_file('my_image.jpg') # encoded the loaded image into a feature vector image_to_be_matched_encoded = face_recognition.face_encodings( image_to_be_matched) # iterate over each image for image in images: # load the image current_image = face_recognition.load_image_file("images/" + image) # encode the loaded image into a feature vector current_image_encoded = face_recognition.face_encodings(current_image) # match your image with the image and check if it matches result = face_recognition.compare_faces( [image_to_be_matched_encoded] current_image_encoded) # check if it was a match if result == True: print "Matched: " + image else: print "Not matched: " + image This Python library encapsulates creating vectors out of faces and differentiating across faces too. References BIBLIOGRAPHY Garvie C. & Frankle J. (2016 April 7). Facial-Recognition Software Might Have a Racial Bias Problem. Retrieved from www.theatlantic.com: https://www.theatlantic.com/technology/archive/2016/04/the-underlying-bias-of-facial-recognition-systems/476991/ [...]
As a financial institution regulated by the FCA, Revolut has the obligation to verify the identity of all customers who want to open a Revolut account. Each prospective customer has to go through a Know Your Customer (KYC) process by submitting a government-issued photo ID and a facial picture of themselves to our partner, Veritas. Veritas then would perform 2 checks: • Document check: To verify that the photo ID is valid and authentic; • Facial Similarity check: To verify that the face in the picture is the same with that on the submitted ID. The customer will ‘pass’ the KYC process and get onboarded if the results of both Document and Facial Similarity checks are ‘clear’. If the result of any check is not ‘clear’, the customer has to submit all the photos again. The “pass rate” is defined as the number of customers who pass both the KYC process divided by the number of customers who attempt the process. Each customer has up to 2 attempts. The pass rate has decreased substantially in the recent period. Please write a report that outlines the root causes and suggest solutions. Relevant files: • Reports of all Facial Similarity checks • Reports of all Document checks • veritas.html - The API documentation of Veritas explaining some terms used in the reports. The candidate is free to use Excel or any scripting language to parse and analyse the data. Please show all your work (including your code if applicable) and assumptions as well as provide a pdf / keynote with your findings (outcomes).