Hello Sandeep, I will try to answer your questions one by one.

1. Generally, instead of using one-class or multi-class classification, feature matching approach is taken for biometric recognition. That means, for each user a set of features are extracted. For testing, similar features are extracted and distance from existing features are measured. The one with the least distance is used to identify a person.

2. For feature matching, whenever a new user is added, their features are added with the corresponding label (person id) in the existing feature set.

Note: a min threshold value is usually required to identify unregistered or unauthorized persons.

3. I think, answer to this is already explained.

Thanks, Jayeeta.

Feature matching is one of the process in a Classification model design.

Similarly, distance based measurement like DTW is a trivial method to determine the Match between input query and stored template of a user.

Block diagrams of enrollment, verification, and identification tasks.

Credits: An Introduction to Biometric Recognition

Anil K. Jain, Fellow, IEEE, Arun Ross, Member, IEEE, and Salil Prabhakar, Member, IEEE

My opinion: when the task to verify, one-class classification model is more justifiable solution for applications, which are rendering services to multiple clients.

With multi-class classification model one can achieve 100% accuracy.

Be it multi-class authentication model or single-class authentication model, the data has to be labelled.

Even object recognition in a group or medical imaging or satellite images or any images, the data is annotated.

Article Continuous Authentication Using One-class Classifiers and their Fusion