News

How to correctly evaluate the effects of Anti-infrared Facial Recognition Camera/ CCTV Privacy Glasses

Facial Image Matching and Recognition
The search matching is carried out for the extracted characteristic data of facial image and characteristic template stored in the database. A threshold value is set. If similarity exceeds such threshold value, the matching results are output. As for the facial recognition, the to-be-recognized facial characteristics and obtained facial characteristic template are compared. Based on similarity, the facial identity information is judged. This process is also divided into two classes: one is confirmation, which is an one-to-one image comparison process, and the other is identification, which is an one-to-many image matching comparison process. The core indexes are matched and recognized:
1）False Accept RATE
It is the misrecognition probability. The facial comparison is passed, but recognizes the wrong person. “False” is recognized as “true”. It is like a person mistaking another person.
2）Pass Rate
It is the correct recognition and pass probability.
False Reject Rate, which is =1- Pass Rate (or FRR for short), is the opposite to the Pass Rate. False Reject Rate means the probability that the database recognizes you but the facial recognition system does not. It is like a person who knows you but does not confirm your identity and says you are not you.
3）Similarity
Refers to the similarity ration between the detected face and the face in the target library in facial comparison. It can be expressed by percentage. For example, the similarity between snapshot face and face in the target library is 80%, it means that a person you see is 80% like the person in your mind.
4）Threshold Value
The aforementioned false accept rate, pass rate, false reject rate and similarity need to be combined with the “Threshold Value” to give the facial recognition system a reasonable evaluation for two reasons:
First, the false accept rate, pass rate and threshold value are related. The higher the threshold value, the lower the pass rate and false accept rate and vice versa. There is a common misunderstanding that the higher the pass rate, the better. Indeed, if the pass rate is higher, the customer experience would be better. However, such statement has two mistakes. One is that the pass rate and false accept rate need to adjust by the threshold value. It is meaningless to mention only pass rate and false accept rate. The second is that false accept rate and pass rate are mutually restricted, and we should not pursue an excellent one-sided data. The other side of high pass rate is high false accept rate, and the anti-risk capacity will decline. Therefore, customer experience and anti-risk capacity (two factors) should be comprehensively considered to determine threshold value and corresponding false accept rate and pass rate. In a more accurate statement, under the designated threshold value, false accept rate of facial recognition system A is lower than that of other systems, and pass rate is higher than that of other systems. This shows that the facial recognition algorithm of facial recognition system A is good.
Second, false accept rate, pass rate and threshold value belong to the non-linear relationship. If threshold value raised constantly, pass rate and false accept rate decline sharply. This shows that threshold value should not be too high. However, as long as threshold value is large enough, the false accept will inevitably appear.
Vivo Detection: One of the technologies that Anti-infrared Facial Recognition Camera/CCTV Privacy Glasses targets
In some identity authentication scenarios, the method for determining the real physiological characteristics of the object can effectively defend the common attack means, such as picture, face change, mask, sheltering and screen re-shooting, so as to help screen fraudulent behavior.
There are three kinds of vivo detection at present, which is mainly based on infrared detection technology. The anti-fake level of living body from low to high is: cooperative vivo detection, silent vivo detection and binocular living body anti-fake detection.

Cooperative Vivo Detection

The most common vivo detection way. Through blink, open mouth, shake one’s head, nod and other cooperative combined actions, whether the subject is the real living body personal operation is verified by facial key point positioning and facial tracking and other technologies.

Silent Vivo Detection

The subject is not required to carry out tedious facial actions but to take a picture or film a facial video in a real-time manner, and then the real person vivo detection can be done. As for the subject, through the facial video played on the displayer, the strict check and recognition can be conducted, and the video playback attack is prevented.

Binocular Vivo Anti-fake Detection

It is the “visible light + near infrared” photoelectric integrated facial vivo detection technology. Its principle lies in the analysis and classification of spectrum information reflected by facial skin under different light conditions. The heterogeneous facial image is related and judged, and the difference between real facial skin and all other attack materials is effectively distinguished. The visible light technology can realize the quick facial recognition. The near infrared imaging technology has such features as “insensitive to illumination, electronic screen imaging failure and imaging while penetrating the sunglasses”. In the actual application scenarios, the malicious intentions of fabricating or stealing the biological characteristics of others for identity authentication can be prevented. The attacks by various means, such as picture, video and 3D mask, can be more effectively prevented. The security of remote verification of identity authenticity is improved.