| CPC G06V 40/172 (2022.01) [G06V 10/75 (2022.01); G06V 10/774 (2022.01)] | 11 Claims |
|
1. A facial recognition method, comprising:
obtaining a to-be-recognized facial image in a target scenario, the target scenario being one of liveness risk detection, hijacking risk detection, and recognition risk detection;
using at least one auxiliary detection model to verify the to-be-recognized facial image, to obtain at least one verification score, wherein the at least one auxiliary detection model include a quality detection model, a liveness detection model, a non-synthetic image detection model;
when each of the at least one verification score is greater than a first decision-making threshold of a corresponding auxiliary detection model in the target scenario, using a feature-matching model to perform feature matching on the to-be-recognized facial image, to obtain a target matching score; and
when the target matching score is greater than a second decision-making threshold of the feature-matching model in the target scenario, determining that the to-be-recognized facial image is successfully recognized, wherein the second decision-making threshold is obtained through joint testing of the feature-matching model and the at least one auxiliary detection model,
wherein:
one of the at least one auxiliary detection model is used as a basic detection model and other ones of the one auxiliary detection model are used as at least one feature verification model;
the first decision-making threshold corresponding to the basic detection model is obtained through testing the basic detection model, and the first decision-making threshold corresponding to each of the at least one feature verification model is obtained through joint testing of the basic detection model and the respective feature verification model; and
the joint testing of the basic detection model and one of the at least one feature verification model comprises:
obtaining a first sample image set and a plurality of sample thresholds;
using the first sample image set to jointly test the basic detection model and the feature verification model, to obtain an interception rate of the feature verification model corresponding to each of the plurality of sample thresholds; and
selecting, from the obtained interception rates, a sample threshold corresponding to a target preset interception rate as a first decision-making threshold corresponding to the feature verification model, wherein different interception rates are obtained for the feature verification model at different model risk levels so that the feature verification model has different first decision-making thresholds to be correspondingly used in different target scenarios.
|