Face Recognition with Anti-Spoof (CNN) on AWS

Business Problem :

Misselling is a major cause of concern where the customer claims he/she did not buy but someone else made the purchase in their name.
Pre conversion verification is a mandatory process where a new customer records a video to say he/she made the purchase. But this process is not full proof and manual checking with an increasing customer base is not prossible.

Solution :

• Developed a Face recognition system to reduce fraud in virtual customer onboarding.

• Developed a face-matching algorithm that compared faces of customers from the frames of mandatory video with their KYC documents submitted at the time of application.

• Face matching uses face embeddings from DeepFace (facebook) model with the help of haar cascade , dlib and opencv.

• Trained a CNN model for anti-spoofing and deployed the model as a service using Docker and ECS.

• Designed the serverless backed on AWS along with services for the admin console to audit the model performance.

---

In case you want to discuss over this project for more details, lets connect!