 In this talk, I will talk with me and my team at Carnegie Mellon can infer from just looking at a small section of your face. In fact, we will show we can perform reliable phase recognition where many current systems will simply fail. Phase recognition and eyes recognition have always made a huge splash in Hollywood movies such as Mission Impossible and Minority Report. Here you see Tom Cruise entering the gap and being identified through his iris and getting personalized ads. You may think this is Hollywood science fiction, but this is actually science fact. This is the technology we have built here in my lab at Carnegie Mellon. This is our system. It can capture someone's iris up to 12 meters away. There is no X marks to spot. It will detect, track the person and capture the iris even if they're wearing glasses. We can even capture someone's iris through the side diameter of a car and make an identification in just a matter of seconds. So how does all this work? First, we have to locate the face in the image and then we detect the eye so we can segment out the iris. The iris is the annual region of the eye between the white sclera and the black pupil. It is the muscle that dilates and contracts the pupil to allow more or less light in depending on how bright the environment is. So we take this circular iris pattern and we unwrap it into a rectangular form. This form is then processed by special functions to generate a series of bits, ones and zeros, to form what we call the iris code of a person. If the irises have iris codes that match with a certain number of bits being the same, then we can say that they belong to the same person. Now, even if two faces look exactly the same as in the case of identical twins, they actually have distinct iris patterns. Actually, an iris pattern of a person is thought to be stable over a person's lifetime, i.e. the pattern of a baby will actually not change as they mature to old age. Now, iris recognition became popular when a famous natural geographic photographer, Steve McCurry, took a picture of this beautiful Afghan girl with amazing green eyes. 17 years later, he wondered what happened to her. Well, they found her and they matched her based on an iris pattern, even though 17 years of hardship passed. So as you can see, iris recognition can play a significant positive role in society. You can imagine our system being used to identify and find missing or abducted children before they're trafficked outside the country. Unlike fingerprint recognition that requires you to touch a physical sensor, iris recognition can be formed at a distance in an unconstrained fashion. But what can we do when we don't have such long-range optics? We have to revert to using the current surveillance infrastructure of the current camera systems. And that's where we come in. We develop smart algorithms that can infer from these low-resolution images. And that's actually what we did in the Boston Marathon case. We reconstructed what this aspect looks like amazingly well. So let us look at what kind of cameras are out there. The surveillance cameras are everywhere. Airports, subways, 7-Eleven, retail stores, even your home. We have to develop smart AI technology that can perform reliable face recognition despite these low-quality images. This is our smart AI face detector. It can actually find faces that are masked. Extreme off-angle, headgear, extreme low-resolution. It can even find a face of a football player playing in the snow with a helmet. So the next step is to be able to find a fine set of maps of the face. We can extract a three-dimensional model of a face from just a single 2D photo. And that's the key thing to be able to perform face recognition. We will show in a second we actually use a single photo of Angelina Jolie and we can render what she looks like at any angle. But what about faces that are masked or occluded? What can we do there? Well, we've developed smart AI software that can infer what the person looks like as a whole face just using this periocular eye region. We're showing here successful that a system can match someone out of a database of 100,000 records. Let us examine how our algorithm is working. The first row is the input to our algorithm, just a periocular region. The second row is the results of our algorithm, the whole face. And the third row is the original faces. So you can compare the second row to the third row. As you can see, our algorithm is performing remarkably. It can actually even neutralize out facial expression and facial hair. So what does this mean for you, the everyday consumer? Our smart AI software is revolutionizing the tech industry. You can imagine a smart home security system running our video analytics performing face recognition to identify intruders who are possibly a burglar and give you a notification before it's too late.