In the field of biometrics, both verification (1:1) and identification (1:N) are important techniques. However, there are some key differences between the two, and the choice depends on the specific application of face recognition.Â
Before we look at the significance of 1:N over 1:1 types of #facerecognition, let us understand the different types of face datasets and the scenarios of obtaining and using them. In the context of face recognition, the subjects can be broadly categorized into two types – Cooperative and Non-Cooperative.
Cooperative subjects refer to individuals who willingly present their faces to the camera and cooperate with the image capture process. Non-cooperative subjects, on the other hand, refer to individuals who do not willingly present their face to the camera or who actively try to avoid being captured, making it more difficult to obtain a usable image.Â
These two lead to 4 combinations of scenarios:Â
Cooperative during registration as well as during recognition. An example of this could be face recognition based attendance and access control, where all the employees of an organization are registering their face image and showing their face to a camera for marking attendance or opening the door. Some law enforcement operations like police patrolling and random checks also fall under this category where subjects are randomly checked against a face database of past offenders.Â
Cooperative during registration, but non-cooperative during recognition. This could be a scenario of general surveillance of a sensitive area for identifying offenders registered in the police database. These datasets are built with proper face photos taken cooperatively when the subject was in custody.Â
Non-cooperative photograph in the face dataset, but cooperative during recognition. This is a scenario where an offender's face photo is available from a CCTV video feed from the past, but people are made to cooperate to show their face, like in a frisking process, before entering a sensitive area. Some military search operations are performed in this manner to identify terrorists in a highly guarded, controlled, surveillance environment.Â
Non-cooperative photograph in the face dataset and also non-cooperative during recognition. Obviously, this could be the most difficult of all scenarios of face recognition. Most law enforcement and military search and surveillance operations fall under this category.Â
#facetagr utilizes advanced techniques like image enhancement, simulations, and resilient recognition with occlusions to manage scenarios involving non-cooperative subjects. These techniques help to improve the accuracy of recognition results despite challenges such as motion blur, occlusion, or attempts to manipulate the system.Â
Overall, while cooperative subjects are easier to work with, non-cooperative subjects pose a greater challenge for face recognition applications.Â
Now, let us come back to the significance of 1:N identification over 1:1 verification:
Verification is not possible in scenarios where the subject is non-cooperative during recognition (scenarios 2 and 4 above) as they have to disclose their identity to perform the 1:1 verification process. However, all four scenarios can be managed with 1:N identification process.Â
The 1:N identification process allows for more efficient and scalable identification of individuals. With 1:N identification, there is no need for the user to manually disclose their identity, which can be time-consuming and error-prone. Instead, the system automatically identifies the individual from a large database of enrolled individuals.Â
Although it can be difficult to deal with non-cooperative subjects, it is better to use 1:N identification as it offers significantly more efficient solutions than 1:1 verification in various applications.