Facial-Recognition System Passes Test On Michelangelo's David (arstechnica.com) 21
An anonymous reader quotes a report from Ars Technica: Facial recognition is a common feature for unlocking smartphones and gaming systems, among other uses. But the technology currently relies upon bulky projectors and lenses, hindering its broader application. Scientists have now developed a new facial recognition system that employs flatter, simpler optics that also require less energy, according to a recent paper published in the journal Nano Letters. The team tested their prototype system with a 3D replica of Michelangelo's famous David sculpture and found it recognized the face as well as existing smartphone facial recognition can. [...]
Wen-Chen Hsu, of National Yang Ming Chiao Tung University and the Hon Hai Research Institute in Taiwan, and colleagues turned to ultrathin optical components known as metasurfaces for a potential solution. These metasurfaces can replace bulkier components for modulating light and have proven popular for depth sensors, endoscopes, tomography. and augmented reality systems, among other emerging applications. Hsu et al. built their own depth-sensing facial recognition system incorporating a metasurface hologram in place of the diffractive optical element. They replaced the standard vertical-cavity surface-emitting laser (VCSEL) with a photonic crystal surface-emitting laser (PCSEL). (The structure of photonic crystals is the mechanism behind the bright iridescent colors in butterfly wings or beetle shells.) The PCSEL can generate its own highly collimated light beam, so there was no need for the bulky light guide or collimation lenses used in VCSEL-based dot projector systems.
The team tested their new system on a replica bust of David, and it worked as well as existing smartphone facial recognition, based on comparing the infrared dot patterns to online photos of the statue. They found that their system generated nearly one and a half times more infrared dots (some 45,700) than the standard commercial technology from a device that is 233 times smaller in terms of surface area than the standard dot projector. "It is a compact and cost-effective system, that can be integrated into a single chip using the flip-chip process of PCSEL," the authors wrote. Additionally, "The metasurface enables the generation of customizable and versatile light patterns, expanding the system's applicability." It's more energy-efficient to boot.
Wen-Chen Hsu, of National Yang Ming Chiao Tung University and the Hon Hai Research Institute in Taiwan, and colleagues turned to ultrathin optical components known as metasurfaces for a potential solution. These metasurfaces can replace bulkier components for modulating light and have proven popular for depth sensors, endoscopes, tomography. and augmented reality systems, among other emerging applications. Hsu et al. built their own depth-sensing facial recognition system incorporating a metasurface hologram in place of the diffractive optical element. They replaced the standard vertical-cavity surface-emitting laser (VCSEL) with a photonic crystal surface-emitting laser (PCSEL). (The structure of photonic crystals is the mechanism behind the bright iridescent colors in butterfly wings or beetle shells.) The PCSEL can generate its own highly collimated light beam, so there was no need for the bulky light guide or collimation lenses used in VCSEL-based dot projector systems.
The team tested their new system on a replica bust of David, and it worked as well as existing smartphone facial recognition, based on comparing the infrared dot patterns to online photos of the statue. They found that their system generated nearly one and a half times more infrared dots (some 45,700) than the standard commercial technology from a device that is 233 times smaller in terms of surface area than the standard dot projector. "It is a compact and cost-effective system, that can be integrated into a single chip using the flip-chip process of PCSEL," the authors wrote. Additionally, "The metasurface enables the generation of customizable and versatile light patterns, expanding the system's applicability." It's more energy-efficient to boot.
How did they test it? (Score:4, Interesting)
To test that it works they will need to test it on (sculptures of?) people who look similar to David. Being able to recognize something implies you won't mis-recognize others as the target -- so I hope they tested for that. Furthermore, for it to be practically useful it has to be able to recognize the face regardless of tan or beard/aging/zits not a sculpture that doesn't change its expression or characteristics day to day. Normally I wouldn't ask that question because I'd expect a basic peer review journal to have made sure they did all that due diligence but since incidents like https://www.telegraph.co.uk/ne... [telegraph.co.uk] I've decided that the ridiculously obvious is something we do need to ask. Anyone read the paper?
Re: How did they test it? (Score:1)
Thatâ(TM)s not the part of the David they tested it on.
Re: (Score:2)
According to the abstract of TFP, they project IR dots onto the face and use the pattern to measure depth. As such, skin tone changes should not affect it. Small changes like spots should also not cause it to fail, as it will have some noise tolerance.
Beards though... I suppose if it only looks at the shape of the nose and eyes, it could work. I guess it will be manufacturer dependent if they go that route. The scientists claim it is better than the current iPhone in terms of the number of measurements made
But does it use AI? (Score:1)
If not then it isn't interesting. /s
So bulky (Score:4, Insightful)
Seriously, who writes this stuff?
Re: (Score:3)
If they can project a field of 45,700 dots through a photonic crystal surface-emitting laser (PCSEL) almost like LIDAR, it might open the door to on-phone 3-D photogrammetry.
Right now the best way to 3-D scan an object is to take approximately 100 photos around the object with something like the KIRI app or the equivalent on an Android phone. Not needing a cloud service to process your 100-photo set would be a huge benefit.
Also, if you see an area that is under scanned or has a hole in the data, it should
Re: (Score:2)
this is actually more about depth perception, i.e. reconstructing 3d geometry of objects via imaging. this is ofc useful for face recognition but has many more applications (i don't really know why they stress facial recognition so much, my guess is as proof of concept. mind you, haven't rtfa).
the face recognition on your smartphone just looks for patterns in a flat image, which is enough for the limited application of unlocking your phone with your face at a stable 30cm distance, with a consumer grade micr
Re: (Score:2)
Hopefully it's small enough that they can get rid of the stupid notch, and make laptop bezels thinner.
Actually the laptop thing is a mixed blessing. My new work Dell's bezel is too small to put one of those sliding covers over the camera, so I ended up using black tape (and disabling it in Device Manager for good measure).
Dell's design is absolute genius. They put the power button right next to the delete key. You better type accurately.
This is worthy of a paper? (Score:1)
I accidentally did this often when I was attempting to do a human-facial-recognition model.
It did a great job on paintings (mona lisa parodies) and sculptures; but was only mediocre on faces.
Thanks (Score:2)
You are very wrong. (Score:2)
You are very wrong.
>> the companies were trying to get their facial recognition stuff to NOT work
They actually try to extract the maximal amount of money from you, so their goal is rather to avoid false negatives which are annoying for the user.
They don't care if a photo of Donald trump will unlock your phone or not, or if there is a backdoor in the unlock software.
Wrong vs. Reality. (Score:2)
They don't care if a photo of Donald trump will unlock your phone or not..
*snort* Really??
I'd say the revenue impact would be nothing compared to being indicted for inciting domestic terrorism against every poor victim suffering from long-TDS after pulling a stunt like that.
81 million. C'mon man. Don't be so heartless. That's a lot of damn victims..
Histroically... (Score:1)
.. the issue has never been with recognising European faces....
Because we keep modelling on European faces
So naturally they opted to brag about recognising the most European face they could find....
Get back to me when they've proven fewer faults than existing non-solutions
iPhone LiDAR (Score:2)
So they could read his WhatsApp messages? (Score:2)
Talking shit to Goliath?
Did it really pass? (Score:3)
So a facial recognition device was pointed at a marble statue and recognized it as a face? I may be mistaken, but faces aren't usually made of marble and a system whose sole job is to recognize HUMAN faces fails when it recognizes anything else as a face. This is especially true for an inanimate object.
Terrible headline (Score:2)
The work involves an improved depth sensor
Did anybody actually read the article before commenting?