Heeseung Choi

Bio: Heeseung Choi is an academic researcher from Korea Institute of Science and Technology. The author has contributed to research in topics: Facial recognition system & Computer science. The author has an hindex of 8, co-authored 28 publications receiving 206 citations.

GAN-Based Anomaly Detection and Localization of Multivariate Time Series Data for Power Plant

Abstract: Recently, as real-time sensor data collection increases in various fields such as power plants, smart factories, and health care systems, anomaly detection for multivariate time series data analysis becomes more important. However, extracting significant features from multivariate time series data is still challenging because it simultaneously takes into account the correlation between the pair of sensors and temporal information of each time series. Meanwhile, in the field of image based anomaly detection, Generative Adversarial Networks(GANs) is developed due to its ability to model the complex high-dimensional distribution of images. In this paper, we propose a novel GAN-based anomaly detection and localization framework along with a transformation method for time series imaging, called distance image. Our goal is to learn a mapping a series of distance image to the next distance image. The transforming multivariate time series into 2D image allows us to exploit encoder and decoder structure. Especially, the encoder with pointwise convolution in a series of images ensures to encode temporal information of each time series data as well the correlation between each variable. As a result, an anomaly can be detected and localized by conducting a residual image and an anomaly score function. We empirically demonstrate the effectiveness of our approach for anomaly detection tasks on a real-world power plant data.

Partial fingerprint matching using minutiae and ridge shape features for small fingerprint scanners

Abstract: New partial fingerprint-matching for small sensors in mobile devices is proposed.The method incorporates new ridge shape features (RSFs) in addition to minutiae.RSFs represent small ridge segments where specific edge shapes are observed.These edge shapes are detectable in conventional 500 dpi images of small sensors. Currently, most mobile devices adopt very small fingerprint sensors that only capture small partial fingerprint images. Accordingly, conventional minutiae-based fingerprint matchers are not capable of providing convincing results due to the insufficiency of minutiae. To secure diverse mobile applications such as those requiring privacy protection and mobile payments, a more accurate fingerprint matcher is demanded. This manuscript proposes a new partial fingerprint-matching method incorporating new ridge shape features (RSFs) in addition to the conventional minutia features. These new RSFs represent the small ridge segments where specific edge shapes (concave and convex) are observed, and they are detectable in conventional 500 dpi images. The RSFs are effectively utilized in the proposed matching scheme which consists of minutiae matching and ridge-feature-matching stages. In the minutiae matching stage, corresponding minutia pairs are determined by comparing the local RSFs and minutiae adjacent to each minutia. During the subsequent ridge-feature-matching stage, the RSFs in the overlapped area of two images are further compared to enhance the matching accuracy. A final matching score is obtained by combining the resulting scores from the two matching stages. Various tests for partial matching were conducted on the FVC2002, FVC2004 and BERC (self-constructed) databases, and the proposed method shows significantly lower equal-error rates compared to other matching methods. The results show that the proposed method improves the accuracy of fingerprint recognition, especially for implementation in mobile devices where small fingerprint scanners are adopted.

Single-view-based 3D facial reconstruction method robust against pose variations

Abstract: The 3D Morphable Model (3DMM) and the Structure from Motion (SfM) methods are widely used for 3D facial reconstruction from 2D single-view or multiple-view images. However, model-based methods suffer from disadvantages such as high computational costs and vulnerability to local minima and head pose variations. The SfM-based methods require multiple facial images in various poses. To overcome these disadvantages, we propose a single-view-based 3D facial reconstruction method that is person-specific and robust to pose variations. Our proposed method combines the simplified 3DMM and the SfM methods. First, 2D initial frontal Facial Feature Points (FFPs) are estimated from a preliminary 3D facial image that is reconstructed by the simplified 3DMM. Second, a bilateral symmetric facial image and its corresponding FFPs are obtained from the original side-view image and corresponding FFPs by using the mirroring technique. Finally, a more accurate the 3D facial shape is reconstructed by the SfM using the frontal, original, and bilateral symmetric FFPs. We evaluated the proposed method using facial images in 35 different poses. The reconstructed facial images and the ground-truth 3D facial shapes obtained from the scanner were compared. The proposed method proved more robust to pose variations than 3DMM. The average 3D Root Mean Square Error (RMSE) between the reconstructed and ground-truth 3D faces was less than 2.6mm when 2D FFPs were manually annotated, and less than 3.5mm when automatically annotated. We propose a single-view-based 3D facial reconstruction method.More robust against pose variations than the previous model-based methods.Person-specific and not biased toward a mean face.

An incremental learning method for spoof fingerprint detection

Abstract: Spoof fingerprint detectors based on static features are built by learning a set of live and fake fingerprint images These learning-based spoof detectors cannot accurately classify new or untrained types of fakes To handle this problem, the existing spoof detector should be incrementally trained on the new types of fakes This paper proposes a new spoof detection framework to learn new types of fakes incrementally without retraining the existing spoof detector repeatedly The proposed model discriminates the newly learned fakes without serious loss of performance for the previously learned fakes and at the same time provides promising detection results for the various types of fakes The proposed spoof detector integrates multiple “experts,” each of which shares the same structure but is separately trained for a different set of fake fingerprints To detect a new type of fake fingerprint, a new expert exclusively trained on the new fake type is integrated into the spoof detector Each expert consists of multiple support vector machines (SVMs) applied by an incremental learning algorithm (Learn++NC), where each SVM adopts one of three texture features for spoof detection Experimental results show the superiority of the proposed method compared with other methods in various scenarios

Query-Based Video Synopsis for Intelligent Traffic Monitoring Applications

Abstract: Synopsis of a long-duration video has many applications in intelligent transportation systems. It can help to monitor traffic with lesser manpower. However, generating meaningful synopsis of a long-duration video recording can be challenging. Often summarized outputs include redundant contents or activities that may not be helpful to the observer. Moving object trajectories are possible sources of information that can be used to generate the synopsis of long-duration videos. The synopsis generation faces challenges due to object tracking, grouping of the trajectories with respect to activity type, object category, and contextual information, and generating smooth synopsis according to a query. In this paper, we propose a method to generate meaningful and smooth synopsis of long-duration videos according to the users’ query. We have tracked moving objects and adopted deep learning to classify the objects into known categories (e.g., car, bike, and pedestrians). We then identify regions in the surveillance scene with the help of unsupervised clustering. Each tube (spatiotemporal object trajectory) is represented by the source and the destination. In the final stage, we take a query from the user and generate the synopsis video by smoothly blending the appropriate tubes over the background frame through energy minimization. The proposed method has been evaluated on two publicly available datasets and our own surveillance datasets. We have compared the method with popular state-of-the-art techniques. The experiments reveal that the proposed method is superior to the existing techniques and it produces visually seamless video synopsis.

A Comprehensive Survey on Pose-Invariant Face Recognition

Abstract: The capacity to recognize faces under varied poses is a fundamental human ability that presents a unique challenge for computer vision systems. Compared to frontal face recognition, which has been intensively studied and has gradually matured in the past few decades, Pose-Invariant Face Recognition (PIFR) remains a largely unsolved problem. However, PIFR is crucial to realizing the full potential of face recognition for real-world applications, since face recognition is intrinsically a passive biometric technology for recognizing uncooperative subjects. In this article, we discuss the inherent difficulties in PIFR and present a comprehensive review of established techniques. Existing PIFR methods can be grouped into four categories, that is, pose-robust feature extraction approaches, multiview subspace learning approaches, face synthesis approaches, and hybrid approaches. The motivations, strategies, pros/cons, and performance of representative approaches are described and compared. Moreover, promising directions for future research are discussed.

A Comprehensive Survey on Pose-Invariant Face Recognition

Abstract: The capacity to recognize faces under varied poses is a fundamental human ability that presents a unique challenge for computer vision systems. Compared to frontal face recognition, which has been intensively studied and has gradually matured in the past few decades, pose-invariant face recognition (PIFR) remains a largely unsolved problem. However, PIFR is crucial to realizing the full potential of face recognition for real-world applications, since face recognition is intrinsically a passive biometric technology for recognizing uncooperative subjects. In this paper, we discuss the inherent difficulties in PIFR and present a comprehensive review of established techniques. Existing PIFR methods can be grouped into four categories, i.e., pose-robust feature extraction approaches, multi-view subspace learning approaches, face synthesis approaches, and hybrid approaches. The motivations, strategies, pros/cons, and performance of representative approaches are described and compared. Moreover, promising directions for future research are discussed.