In this week’s NUS Tech feature, we speak to Dr. Terence Sim, an Associate Professor and avid Peanuts cartoon lover from the National University of Singapore(NUS) School of Computing. Read on as we uncover some of his most interesting work, such as a Motion-based Facial Recognition system and a software tool that recommends and transfers makeup to your face!

Dr. Terence Sim in his office at the NUS School of Computing.

Who are you and what do you do?

Hi, I’m Terence Sim, an Associate Professor at the NUS School of Computing. I carry out research on Image Processing, Computer Vision and Computational Photography. For example, I can analyse facial features for face recognition, transfer makeup to your face using software and enhance the images that you take from your smartphones or cameras.

There are many phones with facial recognition functions nowadays. Exactly how up-to-standard are these technologies?

For normal applications and uses, it is acceptable. However, for high-security purposes, it may fall short because error rates are still high. In other words, there will be situations when it will fail because it cannot recognize your face properly. For example, you might be rejected access to your ibanking account although you are a correctly registered user. When it happens, it can be frustrating.

Another error is called False Acceptance, which happens when someone impersonates you and the system accepts him/her. Imagine a criminal pretending to be you and accessing your bank account. That is why for certain cases, face recognition is not up to not the par yet.

What must be done for face recognition to reach an acceptable range?

For a more accurate recognition system, it is better to make use of multiple biometrics. For instance, you can combine a person’s face with his/her voice. Speaking with your voice is also useful for identifying who you are. When you combine two types of biometrics, it improves accuracy and makes it harder for someone to impersonate you because now the bad guy has to impersonate two of your biometrics.

Do you think that facial recognition is a good form of security?

It is a very convenient method of security. In one photograph, you can capture and recognise many people at once. It is also convenient for mass screening. Imagine the police are looking at a watch list of criminals and are watching people entering a stadium, say for a soccer match. If you put a camera at the gate, you can do mass screening so face recognition is useful for those scenarios.

However, we need to look beyond using the static face. Making use of videos and combining different biometrics are much more secure. One technology that can potentially be used is Infrared. Using infrared light, you can see the vein patterns underneath the skin. Technology companies such as Fujitsu and Hitachi have already developed a palm vein detector which analyses the vein patterns in your palm or finger. It is pretty accurate and very hard to spoof. I mean, how do I spoof your vein patterns?

People tend to alter their physical appearance overtime or their physical features might change as they age. Do these changes affect facial recognition significantly?

Absolutely, it can actually foil your face recognition. If you grow a beard or moustache, that could already cause you to be misrecognized. Certainly, when you age, you can gain or lose weight and develop wrinkles. Such changes can already fool a face recognition system. That is why algorithms need to keep improving.

This is where my work on motion recognition comes in. Using motion is more robust compared to relying just on physical features alone. For instance, we can recognise you based on your smile. People have unique ways of facial movement. This method is good because even if you paint your face, we can still recognise you based on how you smile. Although it is not as accurate as still-face recognition, it is a good complement that improves accuracy.

Combining Appearance and Smile Dynamics in Motion-based Facial Recognition.

What are some of the research you are doing and looking to commercialize?

In terms of commercial potential, I do have a few things that I’m looking for someone to run with it. For example, I have developed a software programme with the ability to analyze a face and recommend suitable makeup based on face shape and skin tone. It might be useful for women who are wondering what kind of makeup will look good on them or if they want to look like a particular celebrity. This could be developed into a useful app or software that could be commercialised.

A software tool that recommends makeup based on a woman’s face shape and skin tone

The other one is makeup compliance. Imagine that you are an air stewardess from Singapore Airlines and you want to check if you comply with the company regulations for hairstyle and makeup. I can come up with the technology, in the form of an app. The flight stewardess can just take a selfie and the app will tell you if you passed based on a set of guidelines. For example, ” Your hair is great, your makeup is not so great. These are the areas you need to make changes to.”

How can you be contacted?

My name is Terence Sim. You can do a Google Search on me or you can just drop me an email at tsim@comp.nus.edu.sg and that would reach me. Thank you!

Watch the video interview with Dr. Terence Sim here:

BLOCK71 Tech Talk: Facial Recognition & Digital Makeup

In this week’s NUS Tech feature, we speak to Dr. Terence Sim, an Associate Professor and Peanuts comic lover from the National University of Singapore(NUS) School of Computing.Watch as we uncover some of his most exciting work: A Motion-based Facial Recognition system that can differentiate identical twins and a software tool that recommends and transfers makeup to your face 👩🏻💄. Looking to develop a Digital Makeup/ Beauty app?Contact Dr. Terence Sim at tsim@comp.nus.edu.sg or learn more about his work at https://tsim49.wixsite.com/terencesim.

Posted by BLOCK71 Singapore on Thursday, 6 June 2019


For startup updates straight to your inbox, subscribe to our weekly newsletter here.