Publications

2024

1. 

   A New Wave of Texture Feature: Enhancing Deepfake Detection via Image Waveform

   Ga San Jhun ,  Pyo Min Hong ,  Keun Lee ,  Jong Hyun Ahn ,  Young Seon Kim ,  Dongwoo Kang , and 1 more author

   In 2024 15th International Conference on Information and Communication Technology Convergence (ICTC) , 2024

   Best paper award

   Abstract Paper

   To address the security threats posed by highly realistic deepfakes, various deepfake detection methods have been proposed. In particular, texture features resulting from the post-processing steps in deepfake generation, such as color correction and edge smoothing have been widely utilized in deepfake detection methods. However their detection method is limited against advanced image synthesis mechanisms. In this paper, we propose a novel deepfake detection method based on texture features, an Image Waveform. Our proposed method converts the pixel values of an image into an Image Waveform to enhance texture features and utilizes two-stream CNN to maximize deepfake detection performance by effectively leveraging the features of the original image. Experimental results of real-world deepfake dataset, confirmed that our proposed method achieves higher detection accuracy compared to existing texture feature-based detection methods.

2. 

   Attention Map Is All We Need for Lightweight Fingerprint Liveness Detection

   Hyun Jun Yook ,  Pyo Min Hong ,  So Hyun Kang ,  Ga San Jhun ,  Jae Eun Seo ,  and  Youn Kyu Lee

   IEEE Access, 2024

   Abstract Paper

   To counter the security threats posed by presentation attacks on fingerprint authentication systems, various deep learning-based fingerprint liveness detection methods have been proposed. However, existing methods typically require significant computing resources or lengthy detection times to achieve high accuracy, which can limit their use in resource-constrained environments such as on-device applications. In this paper, we propose a novel fingerprint liveness detection method that utilizes a Multi-head Self-Attention mechanism. By focusing on important regions of fingerprint images, our proposed method maximizes detection accuracy while simultaneously minimizing both model size and detection time. Our evaluation on real-world datasets demonstrates that our proposed method achieves detection accuracy comparable to state-of-the-art methods while requiring the smallest model size and the least detection time, confirming that our proposed method is the most efficient liveness detection method available.

3. 

   Orthogonal Transform-Driven Data Augmentation for Limited Gaussian-Tainted Dataset

   Jung Won Yoon ,  Hyun Jun Yook ,  Pyo Min Hong ,  Youn Kyu Lee ,  and  Tae Hyung Kim

   IEEE Access, 2024

   Abstract Paper

   A large amount of data collected from sensors exhibits Gaussian noise characteristics, making denoising and related processing critical. However, data scarcity can lead to overfitting, posing challenges in training deep learning-based denoising methods. While various data augmentation methods have been proposed, they do not provide a means to augment original data to large-scale data while preserving the exact noise distribution. To address this, we introduce a novel data augmentation method for data with additive white Gaussian noise (AWGN). Our method is based on two main premises: first, orthogonal transforms preserve the probability distribution of AWGN; second, the signals we aim to recover generally exhibit smooth characteristics, unlike noise. Building on these premises, we propose adaptive smoothnesspromoting orthogonal transforms for augmenting limited existing data. We evaluated the proposed method in Gaussian denoising tasks with limited data and confirmed that it achieves substantial improvement in deep learning model performance, comparable to those obtained with sufficient data.

4. 

   A Comprehensive Risk Analysis Method for Adversarial Attacks on Biometric Authentication Systems

   Seong Hee Park ,  Soo-Hyun Lee ,  Min Young Lim ,  Pyo Min Hong ,  and  Youn Kyu Lee

   IEEE Access, 2024

   Abstract Paper

   Recent threats to deep learning-based biometric authentication systems stem from adversarial attacks exploiting vulnerabilities in deep learning models. While existing studies extensively analyze the risk of such attacks, they primarily focus on isolated modules (e.g., liveness detectors or identity matchers) or specific adversarial attack types (e.g., evasion and poisoning attacks). In this paper, we introduce a novel approach that comprehensively assesses the risk of adversarial attacks by simulating multiple scenarios within biometric authentication systems. We identify the surfaces susceptible to adversarial attacks within these systems and devise scenarios that reflect the dependencies between modules. Moreover, we establish evaluation metrics to comprehensively assess the risk involved. Through a case study conducted on a real-world face recognition system, we successfully demonstrate the effectiveness of our approach. Our approach facilitates the systematic evaluation of the security of target biometric authentication systems against adversarial attacks. Ultimately, it enables the establishment of robust and proactive defense mechanisms.

5. 

   VELCRO: A visual-based programming tool for effortless deep learning model construction

   Min Young Lim* ,  Seong Hee Park* ,  Soo-Hyun Lee* ,  Jung Won Yoon* ,  Pyo Min Hong* ,  Hwajung Yoo* , and 3 more authors

   SoftwareX, 2024

   Abstract Paper

   The development and deployment of deep learning models are currently limited to expert developers due to the requirement for programming expertise, posing challenges for non-expert or inexperienced developers. In this paper, we present VELCRO, a visual-based programming tool that enables intuitive construction, modification, analysis, and generation of deep learning models for both expert and non-expert developers. VELCRO allows users to understand and modify complex architectures of deep learning models using architecture abstraction, and verify architectures through automated architecture validation. Finally, users can convert the user-configured architecture into deep learning code and deploy it in target environments.

2023

1. 

   Adversarial2Adversarial: Defending against Adversarial Fingerprint Attacks without Clean Images

   Pyo Min Hong ,  So Hyun Kang ,  Jinhyeon Kim ,  Ji Hoo Kim ,  and  Youn Kyu Lee

   In 2023 14th International Conference on Information and Communication Technology Convergence (ICTC) , 2023

   Abstract Paper

   A number of denoising-based methods have been proposed to defend against adversarial fingerprint attacks. However, these methods inherently rely on having a clean image that corresponds to each adversarial fingerprint image. In this paper, we propose a novel denoising-based defense method without the need for clean fingerprint images. Our approach leverages a Noise2Noise mechanism, which performs denoising based on the noisy dataset. This enables us to effectively eliminate any adversarial noise that may be embedded in fingerprint images without training on clean fingerprint images. The experimental results on real-world datasets confirm that our method is robust against untrained adversarial fingerprint attacks while outperforming existing methods.

2. 

   Two-Stage Image Restoration: A Shield against Adversarial Attacks on Face Identification

   Min Young Lim ,  Pyo Min Hong ,  and  Youn Kyu Lee

   In 2023 14th International Conference on Information and Communication Technology Convergence (ICTC) , 2023

   Abstract Paper

   Various methods have been proposed to defend face identification systems against adversarial attacks by eliminating adversarial perturbations from target face images. However, it is challenging to eliminate such perturbations while preserving the crucial facial features within the images. In this paper, we propose a novel purification method for an effective defense against adversarial attacks on target face images. Our method incorporates a two-stage image restoration utilizing diffusion, consisting of sequential super-resolution-based image restoration followed by colorization-based image restoration. The experimental results demonstrate the effectiveness of our method in eliminating perturbations while preserving the identity of the facial features.

3. 

   Defending Against Adversarial Fingerprint Attacks Based on Deep Image Prior

   Hwajung Yoo ,  Pyo Min Hong ,  Taeyong Kim ,  Jung Won Yoon ,  and  Youn Kyu Lee

   IEEE Access, 2023

   Abstract Paper

   Recently, deep learning-based biometric authentication systems, especially fingerprint authentication, have been used widely in real-world. However, these systems are vulnerable to adversarial attacks which prevent deep learning models from distinguishing input data properly. To solve these problems, various defense methods have been proposed, especially utilizing denoising mechanisms, but they provided limited defense performance. In this study, we proposed a new defense method against adversarial fingerprint attacks. To ensure defense performance, we have introduced Deep Image Prior mechanism which has superior performance in image reconstruction without prior training and a large amount of dataset. The proposed method aims to remove adversarial perturbations of the input fingerprint image and reconstruct it close to the original fingerprint image by adapting Deep Image Prior. Our method has achieved robust defense performance against various types of adversarial fingerprint attacks across different datasets, encompassing variations in sensors, shapes, and materials of fingerprint images. Furthermore, our method has demonstrated that it is superior to other image reconstruction methods.

4. 

   A Study on Adversarial Fingerprint Defense Using Self-Supervised Denoising from Single Image

   Pyo Min Hong ,  Hwajung Yoo ,  Taeyong Kim ,  Jung Won Yoon ,  Tae Hyung Kim ,  and  Youn Kyu Lee

   The Journal of Korean Institute of Communications and Information Sciences, 2023

   Abstract Paper

   Adversarial fingerprint attacks exploit deep learning-based fingerprint authentication systems, causing abnormal behavior in the model. The emergence of various forms of adversarial attacks has created vulnerabilities in deep learning-based fingerprint authentication systems, leading to a new security issue. In this paper, we propose a defense mechanism that provides generalized performance against various adversarial fingerprint attacks without requiring multiple types of fingerprint images. The proposed method utilizes a single image-based self-supervised denoising technique to effectively remove adversarial noise from input fingerprint images while restoring them to their original state, providing robust defense performance against adversarial fingerprint attacks. Furthermore, it offers superior defense performance compared to existing image restoration methods without requiring pre-training on large-scale fingerprint images.

2022

1. 

   A Central Point-based Analysis for Fingerprint Liveness Detection

   Min Young Lim* ,  Tae Yong Kim* ,  Joung Eun Park* ,  Pyo Min Hong* ,  and  Youn Kyu Lee

   In 2022 13th International Conference on Information and Communication Technology Convergence (ICTC) , 2022

   Best paper award

   Abstract Paper

   To minimize security threats using fake fingerprints, various CNN-based fingerprint liveness detection methods have been proposed. However, since existing methods mainly employ a random crop method, key information of target fingerprint may be overlooked during the training process, which may lead to decreased detection accuracy. In this paper, we propose a new fingerprint liveness detection method based on central point analysis of fingerprints. The proposed method measures a central point of target fingerprint, extracts the crops with different sizes based on the central point, and fuses the liveness scores inferred from each crop-size model. As a result of validating our method using real datasets, it was confirmed that our method effectively detects fake fingerprints compared to existing methods.