US-20260127912-A1 - METHOD AND APPARATUS FOR FAKE FINGERPRINTS GENERATION, AND TRAINING METHOD OF ARTIFICIAL INTELLIGENCE FOR IDENTIFYING FAKE FINGERPRINTS

US20260127912A1US 20260127912 A1US20260127912 A1US 20260127912A1US-20260127912-A1

Abstract

There is provided a method for training a fake fingerprint detection model, performed by a training device for the fake fingerprint detection model, the method comprising: acquiring a physical fake fingerprint image generated in a physical manner targeting a finger; providing generation constraints and a unique identification information to a training fingerprint image generation model; generating a training fingerprint image by using the training fingerprint image generation model, so that the unique identification information of the training fingerprint image is mapped onto a real fingerprint of the finger; and training the fake fingerprint detection model using the physical fake fingerprint image and the training fingerprint image in a transfer learning manner or an ensemble learning manner.

Inventors

Beom Jun KIM
Jong Man Lee
Hochul SHIN
Jae Hyun Park
Young Mook KANG

Assignees

SUPREMA INC.

Dates

Publication Date: 20260507
Application Date: 20241212
Priority Date: 20241106

Claims (19)

1 . A method for training a fake fingerprint detection model, performed by a training device for the fake fingerprint detection model, the method comprising: acquiring a physical fake fingerprint image generated in a physical manner targeting a finger; providing generation constraints and a unique identification information to a training fingerprint image generation model; generating a training fingerprint image to be used for training the fake fingerprint detection model by using the training fingerprint image generation model; and training the fake fingerprint detection model using the physical fake fingerprint image and the training fingerprint image in a transfer learning manner or an ensemble learning manner, wherein the training fingerprint image generation model is trained to generate different training fingerprint images based on conditions included in the generation constraints.
2 . The method of claim 1 , wherein in the transfer learning manner, the physical fake fingerprint image is used for pre-training and the training fingerprint image is used for fine-tuning.
3 . The method of claim 2 , wherein in the pre-training, an image with masking applied to a portion of the physical fake fingerprint image is provided as an input for training, and an image with no masking applied is provided as a correct answer for training.
4 . The method of claim 1 , wherein the fake fingerprint detection model includes a physical fake fingerprint detection model and a training fake fingerprint detection model, and wherein in the training of the ensemble learning manner, the physical fake fingerprint detection model and the training fake fingerprint detection model are trained separately, the physical fake fingerprint detection model is trained by receiving the physical fake fingerprint image as an input for training and receiving whether the received image is fake or not as a correct answer for training, and the training fake fingerprint detection model is trained by receiving the training fingerprint image as an input for training and receiving whether the received image is fake or not as a correct answer for training.
5 . The method of claim 1 , wherein the unique identification information includes at least one of first identification information based on start points, end points and branch points of ridges in a fingerprint of the finger, second identification information based on orientations of the ridges, and third identification information based on a distribution density of the ridges.
6 . The method of claim 5 , wherein the generation constraints include information about a sensor that identifies the fingerprint, information about surrounding environment in which the fingerprint is identified, and labeling information about the fingerprint.
7 . The method of claim 6 , wherein the labeling information includes information about a type of physical means used to generate the physical fake fingerprint image.
8 . (canceled)
9 . The method of claim 1 , wherein the training fingerprint image generation model acquires random noise and generates different training fingerprint images according to the random noise even when same generation constraints are applied.
10 . A device for training a fake fingerprint detection model, the device comprising: a memory storing computer-executable instructions; and a processor for executing the instructions to: acquire a physical fake fingerprint image generated in a physical manner targeting a finger; provide generation constraints and a unique identification information to a training fingerprint image generation model; generate a training fingerprint image to be used for training the fake fingerprint detection model by using the training fingerprint image generation model; and train the fake fingerprint detection model using the physical fake fingerprint image and the training fingerprint image in a transfer learning manner or an ensemble learning manner, wherein the training fingerprint image generation model is trained to generate different training fingerprint images based on conditions included in the generation constraints.
11 . The device of claim 10 , wherein in the transfer learning manner, the physical fake fingerprint image is used for pre-training and the training fingerprint image is used for fine-tuning.
12 . The device of claim 11 , wherein in the pre-training, an image with masking applied to a portion of the physical fake fingerprint image is provided as an input for training, and an image with no masking applied is provided as a correct answer for training.
13 . The device of claim 10 , wherein the fake fingerprint detection model includes a physical fake fingerprint detection model and a training fake fingerprint detection model, and wherein in the training of the ensemble learning manner, the physical fake fingerprint detection model and the training fake fingerprint detection model are trained separately, the physical fake fingerprint detection model is trained by receiving the physical fake fingerprint image as an input for training and receiving whether the received image is fake or not as a correct answer for training, and the training fake fingerprint detection model is trained by receiving the training fingerprint image as an input for training and receiving whether the received image is fake or not as a correct answer for training.
14 . The device of claim 10 , wherein the unique identification information includes at least one of first identification information based on start points, end points and branch points of ridges in a fingerprint of the finger, second identification information based on orientations of the ridges, and third identification information based on a distribution density of the ridges.
15 . The device of claim 14 , wherein the generation constraints include information about a sensor that identifies the fingerprint, information about surrounding environment in which the fingerprint is identified, and labeling information about the fingerprint.
16 . The device of claim 15 , wherein the labeling information includes information about a type of physical means used to generate the physical fake fingerprint image.
17 . (canceled)
18 . The device of claim 10 , wherein the training fingerprint image generation model acquires random noise and generates different training fingerprint images according to the random noise even when same generation constraints are applied.
19 . A non-transitory computer-readable recording medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, cause the processor to perform a method for training a fake fingerprint detection model, the method comprising: acquiring a physical fake fingerprint image generated in a physical manner targeting a finger; providing generation constraints and a unique identification information to a training fingerprint image generation model; generating a training fingerprint image to be used for training the fake fingerprint detection model by using the training fingerprint image generation model; and training the fake fingerprint detection model using the physical fake fingerprint image and the training fingerprint image in a transfer learning manner or an ensemble learning manner, wherein the training fingerprint image generation model is trained to generate different training fingerprint images based on conditions included in the generation constraints.

Description

CROSS-REFERENCE TO RELATED APPLICATION This application claims priority to Korean Patent Application No. 10-2024-0156337, filed on Nov. 6, 2024, the entirety of which is incorporated herein by reference for all purposes. TECHNICAL FIELD The present disclosure relates to a method and device for generating fake fingerprints, and a method for training an artificial intelligence model for detecting fake fingerprints. This work was supported by Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (Ministry of Science and ICT) (Project unique No.: 2710007816; Project No.: II201787; R&D project: Information and Communication Broadcasting Innovation Talent Training (R&D); Research Project Title: Development of communication/computing convergence innovation technology for super-intelligent services; and Project period: 2024.01.01.˜2024.12.31.). BACKGROUND Fingerprint recognition is one of the most widely used biometric technologies today, and is rapidly replacing passwords composed of letters or patterns in daily life. Many people perceive fingerprint recognition technology as being more resistant to leakage or duplicate than passwords composed of letters or patterns, however fingerprints are not free from the risk of duplication. Therefore, if a fingerprint is exposed and duplicated by a third party, the duplicated fake fingerprint may be used to obtain authentication. Meanwhile, fingerprints can be easily exposed to the outside world, such as being left on objects touched by the hand, which leads to a security vulnerability. Accordingly, the development of technologies to distinguish between real fingerprints and fake fingerprints (e.g., the development of fake fingerprint detection technology using artificial intelligence) is actively underway. However, there is a limitation in improving the performance of the fake fingerprint detection models due to insufficient training data to train artificial intelligence algorithms used for detect fake fingerprints. SUMMARY In view of the above, the present disclosure provides a method for generating a large amount of training data required to training an artificial intelligence model for detecting fake fingerprints. In addition, the present disclosure provides a method for training a fake fingerprint detection model using training data generated by the aforementioned method. However, the problem to be solved by the present disclosure is not limited to that mentioned above, and other problems to be solved that are not mentioned may be clearly understood by those of ordinary skill in the art to which the present disclosure belongs from the following description. In accordance with an aspect of the present disclosure, there is provided a method for training a fake fingerprint detection model, performed by a training device for the fake fingerprint detection model, the method comprising: acquiring a physical fake fingerprint image generated in a physical manner targeting a finger; providing generation constraints and a unique identification information to a training fingerprint image generation model; generating a training fingerprint image by using the training fingerprint image generation model, so that the unique identification information of the training fingerprint image is mapped onto a real fingerprint of the finger; and training the fake fingerprint detection model using the physical fake fingerprint image and the training fingerprint image in a transfer learning manner or an ensemble learning manner. In the transfer learning manner, the physical fake fingerprint image may be used for pre-training and the training fingerprint image may be used for fine-tuning. In the pre-training, an image with masking applied to a portion of the physical fake fingerprint image may be provided as an input for training, and an image with no masking applied may be provided as a correct answer for training. The fake fingerprint detection model may include a physical fake fingerprint detection model and a training fake fingerprint detection model, and wherein in the training of the ensemble learning manner, the physical fake fingerprint detection model and the training fake fingerprint detection model may be trained separately, the physical fake fingerprint detection model may be trained by receiving the physical fake fingerprint image as an input for training and receiving whether the received image may be fake or not as a correct answer for training, and the training fake fingerprint detection model may be trained by receiving the training fingerprint image as an input for training and receiving whether the received image is fake or not as a correct answer for training. The unique identification information may include at least one of first identification information based on start points, end points and branch points of ridges in the fingerprint of the finger, second identification information based on orientations of the