I think Apple's R&D around facial recognition is not for authentication, but rather for:
Gaming
Augmented Reality
Facetime
... etc.
That is (as @radarthekat mentioned), use cases where the device must detect the EXISTENCE of a face, but not necessarily know the IDENTITY of that face.
I buy this theory more than any other. Replacing the touch ID does not make sense. The finger print is a rock solid biometric to use. Judging by all of the facial recognition I have experienced, no way will it work. I can imagine all of the failed attempts at the cash register. No, I think slprescott has it right.
We are forgetting something important. If I remember correctly, when it came out Apple purchased faceshift. They pointed out this company had developed technology to scan, detect and put faces in a 3D world without actually putting anything on the actual face. Current Cgi and facial recognition requires suits and sensors to accurately accomplish a complete body scan.
Now, imagine Apple getting up on that stage one fall and one of those game companies coming out and holds up an iPhone and Tim Cook scans his face and some orc or other game creature shows up with his facial features... how do you think that's gonna change the playing field? This is what I suspsect apples first step toward augmented reality.
stop and ask yourself, where's the money? The money isn't in facial id security.
The money is in being the ONLY platform to Allow you to use your phone to put your ugly mug on an avatar... primesense was the first purchase... faceshift and emotient and all the others are simply building on the patent trove of primesense, and at the same time defining apples direction. To find the direction you must go to the original purchase.
If Apple is going to put a 3D camera into its phones it's going to be for gaming. Because that's where the money is.
We are forgetting something important. If I remember correctly, when it came out Apple purchased faceshift. They pointed out this company had developed technology to scan, detect and put faces in a 3D world without actually putting anything on the actual face. Current Cgi and facial recognition requires suits and sensors to accurately accomplish a complete body scan.
Now, imagine Apple getting up on that stage one fall and one of those game companies coming out and holds up an iPhone and Tim Cook scans his face and some orc or other game creature shows up with his facial features... how do you think that's gonna change the playing field? This is what I suspsect apples first step toward augmented reality.
stop and ask yourself, where's the money? The money isn't in facial id security.
The money is in being the ONLY platform to Allow you to use your phone to put your ugly mug on an avatar... primesense was the first purchase... faceshift and emotient and all the others are simply building on the patent trove of primesense, and at the same time defining apples direction. To find the direction you must go to the original purchase.
If Apple is going to put a 3D camera into its phones it's going to be for gaming. Because that's where the money is.
They've purchase so many, and there may be more that haven't been reported. Wikipedia lists augmented reality, motion capture, and facial recognition software. Note that PrimeSense is listed as a semiconductor company.
The real reason Facial Recognition won't replace TouchId is that facial recognition, as developed by Emotient and Faceshift, is simply not the technology used for biometrics. It's in the name itself, which I continually try to educate the AI staffers on. Facial recognition, as implied by the word Facial, is recognition of facial features and expressions. Perfect for an application such as Faceshift developed. When you want to map facial features, like smiling, crinkled eyes, eyebrow movements, from a human subject onto an animated avatar inside a game, for example, the technology does not need to identify who you are; it merely needs to map the human facial features it sees onto the avatar. So that's what 'Facial Recognition' does, and no more.
Also mentioned in the article is the term 'face detection.' Again, while this is an important technology on its own, it's not all that is required for biometric identification, though it's often a subset of the process. Face detection is simply the process of identifying, and sometimes tracking, a human face within a scene. It typically uses Fourier transforms and blurring alogorithms to detect patterns of eyes, nose, mouth, even at somewhat oblique angles. But face detection also doesn't specifically identify an individual. That's the job of...
Face Recognition. Face recognition technology often begins with face detection algorithms, so that any faces that appear in a scene, in the view of a video camera, can be detected and captured. Face recognition might then also employ facial recognition algorithms (but not necessarily) to identify specific facial expressions on the detected faces. But then Face Recognition adds a further, crucial step, applying algorithms not needed by Face Detection or Facial Recognition systems to match specific aspects and calculations made against each face against results stored in a database. This allows the face recognition system to perform its primary task, that of biometric identification, a task that is specifically NOT part of the definition of Face Detection or of Facial Recognition.
See why I've been nitpicking the terminology all these past months?
You are correct in all that. A couple of years ago, Microsoft tried a Win Phone with facial recognition for unlocking. It worked about two thirds of the time. But if you did wear glasses, or sunglasses, or a hat, or your beard changed, or any number of things, it didn't work. Sunlight was a problem, shadows were a problem and not enough light was a problem.
the fact that it worked two thirds of the time was an accomplishment, but obviously, that's not nearly enough. It needs to work 99.9% of the time. No doubt it is getting better. But commercial and military systems have very expensive, and big, mechanisms. They also control the lighting. And you're expected to appear the same each time you use it.
its also clumsy.
You are completely mistaken...
The phones you refer to are the Lumia 950 and 950 XL. They do not have face recognition - they have optional iris detection. This was released as an optional beta feature when these phones became available in November of 2015. Even then the iris recognition worked much more than half the time - I know I have both a 950 and 950 XL (the XL is still my daily driver). This has greatly improved since then as well - I would guess that it works about 90 - 95% of the time. And it has become much faster as well.
As for the lighting, again - not so, except for the bright sunlight. That was and is still a problem since the system uses an ir scanner. But, barring standing in direct sun, the system works even in pitch dark. It even works with your glasses on (including non-mirrored sun glasses) - because again it uses and ir scan. So, as long as the ir beam can get to your iris the detection works. I did find it dropped in accuracy with glasses at first - but, taking the option to improve detection and repeating the setup with my glasses on made that a non-issue.
Interestingly, I recently bought a windows hello enabled web cam for my PC. This does in fact use face recognition - and it works almost 100% of the time in all lighting, except in bright sunlight because well, the same reason. It uses an ir sensor. It is very fast and convenient. I sit down, move my mouse and my pc wakes up and logs me in. If it fails to recognize me for some reason (usually during the hour or so a day that I have bright sunlight to my back) I just enter a pin code and I'm in.
The Surface Pro 4 has this feature built in. This is built on the Intel Real Sense camera - which was already shown working in a smart phone.
And before the inevitable non-sense about using pictures to fool Windows Hello face recognition - that can't work. You can't use just any old camera. The system uses ir and the heat map is part of the algorithm. It requires a living 3d face in front of it. And the system is good enough to not even be fooled by identical twins...
I just looked up the false positive rates for touch id and windows hello... Apparently, Apple claims a 1 in 50,000 false positive rate. Windows Hello claims a 1 in 100,000 false positive rate. Hence, face recognition would appear to be twice as good as finger print scanning for security...
I just looked up the false positive rates for touch id and windows hello... Apparently, Apple claims a 1 in 50,000 false positive rate. Windows Hello claims a 1 in 100,000 false positive rate. Hence, face recognition would appear to be twice as good as finger print scanning for security...
Just a note.
Have any of these claims been independently confirmed? Note that all biometrics offer great security. Your iOS keyboard offers over a billion combinations with only 4 characters.
I am afraid that Mikeycampbell81 misses the point that radarthekat made. Accuracy in communications depends on both parties using the same standard of definition of words. A case in point, if you say Kleenex, are you referring to a facial tissue, a particular facial tissue or the company? When I ask for Coke that is not a generic request for cola based soft drink. If I am offered Pepsi I take water, personal preference, but I didn't ask for Pepsi or any generic drink. As words, both spoken and written, are our primary method of communication we need to strive for accuracy.
No misunderstanding here. Apple in its IP uses "facial recognition" interchangeably with "face detection" and "face recognition," so we do the same to avoid confusion for those referencing cited patents. The terms are also not generic trademarks.
With face recognition for unlocking a phone, a cop can take your phone then quickly aim it at your face and voila, phone unlocked without ever touching you. TouchID, if you don't wish to cooperate, somebody will have to physically manhandle you to bring your fingertip to your phone. That's a big divide there in some jurisdictions; allowing the first action but not the second.
How many times have you unlocked an iPhone that was sitting face up on a table next to you? All that was required was to place a finger on the Home button and ta-da! ... the phone is unlocked. Now imagine having to either pick up the phone and point it at your face, or lean over the phone so your face is directly over the screen, looking straight down at it. Inconvenient, to say the least.
I think Apple's R&D around facial recognition is not for authentication, but rather for:
Gaming
Augmented Reality
Facetime
... etc.
That is (as @radarthekat mentioned), use cases where the device must detect the EXISTENCE of a face, but not necessarily know the IDENTITY of that face.
Here's a fun idea for an app developer: While texting, instead of searching for the right emoji for your expression, use facial recognition to scan whatever face you want to make, turn that into a static or animated avatar emoji and voila! Perhaps it'd be relatively easy to build with a FacialRecognition Kit for devs.
I am afraid that Mikeycampbell81 misses the point that radarthekat made. Accuracy in communications depends on both parties using the same standard of definition of words. A case in point, if you say Kleenex, are you referring to a facial tissue, a particular facial tissue or the company? When I ask for Coke that is not a generic request for cola based soft drink. If I am offered Pepsi I take water, personal preference, but I didn't ask for Pepsi or any generic drink. As words, both spoken and written, are our primary method of communication we need to strive for accuracy.
No misunderstanding here. Apple in its IP uses "facial recognition" interchangeably with "face detection" and "face recognition," so we do the same to avoid confusion for those referencing cited patents. The terms are also not generic trademarks.
They don't actually use the terms interchangeably. They use all these terms, but they use them each where each is appropriate. As I mentioned, face detection and facial recognition can be steps in the face recognition process, and therefore may be mentioned within a patent's description and claims, even if that patent is focused on biometric authentication. Another read of the patents that have been referenced here in previous articles, where we've had this discussion, will be informative and I think you'll find my use of the terminology to hold true.
The real reason Facial Recognition won't replace TouchId is that facial recognition, as developed by Emotient and Faceshift, is simply not the technology used for biometrics. It's in the name itself, which I continually try to educate the AI staffers on. Facial recognition, as implied by the word Facial, is recognition of facial features and expressions. Perfect for an application such as Faceshift developed. When you want to map facial features, like smiling, crinkled eyes, eyebrow movements, from a human subject onto an animated avatar inside a game, for example, the technology does not need to identify who you are; it merely needs to map the human facial features it sees onto the avatar. So that's what 'Facial Recognition' does, and no more.
Also mentioned in the article is the term 'face detection.' Again, while this is an important technology on its own, it's not all that is required for biometric identification, though it's often a subset of the process. Face detection is simply the process of identifying, and sometimes tracking, a human face within a scene. It typically uses Fourier transforms and blurring alogorithms to detect patterns of eyes, nose, mouth, even at somewhat oblique angles. But face detection also doesn't specifically identify an individual. That's the job of...
Face Recognition. Face recognition technology often begins with face detection algorithms, so that any faces that appear in a scene, in the view of a video camera, can be detected and captured. Face recognition might then also employ facial recognition algorithms (but not necessarily) to identify specific facial expressions on the detected faces. But then Face Recognition adds a further, crucial step, applying algorithms not needed by Face Detection or Facial Recognition systems to match specific aspects and calculations made against each face against results stored in a database. This allows the face recognition system to perform its primary task, that of biometric identification, a task that is specifically NOT part of the definition of Face Detection or of Facial Recognition.
See why I've been nitpicking the terminology all these past months?
You are correct in all that. A couple of years ago, Microsoft tried a Win Phone with facial recognition for unlocking. It worked about two thirds of the time. But if you did wear glasses, or sunglasses, or a hat, or your beard changed, or any number of things, it didn't work. Sunlight was a problem, shadows were a problem and not enough light was a problem.
the fact that it worked two thirds of the time was an accomplishment, but obviously, that's not nearly enough. It needs to work 99.9% of the time. No doubt it is getting better. But commercial and military systems have very expensive, and big, mechanisms. They also control the lighting. And you're expected to appear the same each time you use it.
its also clumsy.
You are completely mistaken...
The phones you refer to are the Lumia 950 and 950 XL. They do not have face recognition - they have optional iris detection. This was released as an optional beta feature when these phones became available in November of 2015. Even then the iris recognition worked much more than half the time - I know I have both a 950 and 950 XL (the XL is still my daily driver). This has greatly improved since then as well - I would guess that it works about 90 - 95% of the time. And it has become much faster as well.
As for the lighting, again - not so, except for the bright sunlight. That was and is still a problem since the system uses an ir scanner. But, barring standing in direct sun, the system works even in pitch dark. It even works with your glasses on (including non-mirrored sun glasses) - because again it uses and ir scan. So, as long as the ir beam can get to your iris the detection works. I did find it dropped in accuracy with glasses at first - but, taking the option to improve detection and repeating the setup with my glasses on made that a non-issue.
Interestingly, I recently bought a windows hello enabled web cam for my PC. This does in fact use face recognition - and it works almost 100% of the time in all lighting, except in bright sunlight because well, the same reason. It uses an ir sensor. It is very fast and convenient. I sit down, move my mouse and my pc wakes up and logs me in. If it fails to recognize me for some reason (usually during the hour or so a day that I have bright sunlight to my back) I just enter a pin code and I'm in.
The Surface Pro 4 has this feature built in. This is built on the Intel Real Sense camera - which was already shown working in a smart phone.
And before the inevitable non-sense about using pictures to fool Windows Hello face recognition - that can't work. You can't use just any old camera. The system uses ir and the heat map is part of the algorithm. It requires a living 3d face in front of it. And the system is good enough to not even be fooled by identical twins...
Yes, you're right. It was iris scan. Sorry. It was so long ago, it seems, that I had forgotten.
atill, the rest of my statement stands. This will be no better than the ugly performance of Microsoft's product.
I think Apple's R&D around facial recognition is not for authentication, but rather for:
Gaming
Augmented Reality
Facetime
... etc.
That is (as @radarthekat mentioned), use cases where the device must detect the EXISTENCE of a face, but not necessarily know the IDENTITY of that face.
Here's a fun idea for an app developer: While texting, instead of searching for the right emoji for your expression, use facial recognition to scan whatever face you want to make, turn that into a static or animated avatar emoji and voila! Perhaps it'd be relatively easy to build with a FacialRecognition Kit for devs.
The pile of poop emoji is gonna require an amazing stank face. 💩
Comments
Now, imagine Apple getting up on that stage one fall and one of those game companies coming out and holds up an iPhone and Tim Cook scans his face and some orc or other game creature shows up with his facial features... how do you think that's gonna change the playing field? This is what I suspsect apples first step toward augmented reality.
stop and ask yourself, where's the money? The money isn't in facial id security.
The money is in being the ONLY platform to
Allow you to use your phone to put your ugly mug on an avatar... primesense was the first purchase... faceshift and emotient and all the others are simply building on the patent trove of primesense, and at the same time defining apples direction. To find the direction you must go to the original purchase.
If Apple is going to put a 3D camera into its phones it's going to be for gaming. Because that's where the money is.
You are completely mistaken...
The phones you refer to are the Lumia 950 and 950 XL. They do not have face recognition - they have optional iris detection. This was released as an optional beta feature when these phones became available in November of 2015. Even then the iris recognition worked much more than half the time - I know I have both a 950 and 950 XL (the XL is still my daily driver). This has greatly improved since then as well - I would guess that it works about 90 - 95% of the time. And it has become much faster as well.
As for the lighting, again - not so, except for the bright sunlight. That was and is still a problem since the system uses an ir scanner. But, barring standing in direct sun, the system works even in pitch dark. It even works with your glasses on (including non-mirrored sun glasses) - because again it uses and ir scan. So, as long as the ir beam can get to your iris the detection works. I did find it dropped in accuracy with glasses at first - but, taking the option to improve detection and repeating the setup with my glasses on made that a non-issue.
Interestingly, I recently bought a windows hello enabled web cam for my PC. This does in fact use face recognition - and it works almost 100% of the time in all lighting, except in bright sunlight because well, the same reason. It uses an ir sensor. It is very fast and convenient. I sit down, move my mouse and my pc wakes up and logs me in. If it fails to recognize me for some reason (usually during the hour or so a day that I have bright sunlight to my back) I just enter a pin code and I'm in.
The Surface Pro 4 has this feature built in. This is built on the Intel Real Sense camera - which was already shown working in a smart phone.
And before the inevitable non-sense about using pictures to fool Windows Hello face recognition - that can't work. You can't use just any old camera. The system uses ir and the heat map is part of the algorithm. It requires a living 3d face in front of it. And the system is good enough to not even be fooled by identical twins...
Just a note.
atill, the rest of my statement stands. This will be no better than the ugly performance of Microsoft's product.