Historic Archives
The proposed research project aims to leverage the transformative power of Artificial Intelligence (AI) to manage, analyze, and interpret historical archives from the Chilean dictatorship era. Driven by the critical need to consolidate and scrutinize thousands of documents, old photographs, and audio recordings dispersed across various archives and collections, this project addresses the challenges of fragmented archives –a task nearly impossible to achieve manually. The project plans to digitize, transcribe, and methodically analyze large datasets using advanced AI techniques, including natural language processing (NLP), machine learning (ML), and computer vision (CV). This holistic approach will not only preserve the integrity of historical records but also unveil new perspectives and deeper insights into the socio-political dynamics of the Chilean dictatorship, thus making a substantial contribution to the fields of historical research, education, and transitional justice.
IMFD, Millenium Institute. See more at www.nuestramemorIA.cl
Biometrics
Frontiers of Face Recognition in Low-Quality Images
Our project’s main goal is to investigate the frontiers of human and computer vision working alone and how computer vision and human vision can work in collaboration to recognize face images depending on their low-quality characteristics. Thus, our main hypothesis is that human and computer vision in face recognition of low-quality images can mutually benefit from one another. That means, when recognizing low-quality face images, i) human vision accuracy can be improved if the face images are enhanced using computer vision algorithms before the humans examine the images, and ii) computer vision accuracy can be increased if we include new models based on human perception.
FONDECYT, in collaboration with the School of Psychology of the Catholic University of Chile, the University of Notre Dame, the University of Texas, and BiometryPass Co.
Face Matching from Teenagers to Young Adults
The project is focused on an important topic of current research in the area of facial recognition. The basic problem is developing high-accuracy facial recognition for matching images taken in early teenage years, as early as 12 years old, with images taken as young adults, up to mid or late 20s. Two elements combine to make this particular instance of the face recognition problem quite difficult. i) One element is that a person’s face naturally goes through substantial real changes from early teenage years to middle 20’s. ii) The other element that makes this problem hard is that the initial image is a document image, a face image printed on an identity card. This problem of `Matching ID Document Photos to Selfies’ is a hot topic in face recognition research.
SEED FUND UC-ND, in collaboration with the University of Notre Dame, the University of North Carolina Wilmington, and BiometryPass Co.
Low-Quality Face Recognition in the Wild
Although face recognition systems have achieved impressive performance in recent years, the low-resolution face recognition (LRFR) task remains challenging, especially when the LR faces are captured under non-ideal conditions, as is common in surveillance-based applications. Faces captured in such conditions are often contaminated by blur, nonuniform lighting, and nonfrontal face pose. In this project, we aim to analyze face recognition techniques using data captured under low-quality conditions in the wild. The performance gap between general unconstrained face recognition and face recognition in surveillance videos is shown, and new algorithms have been designed to tackle this specific problem.
In collaboration with the Computer Vision Research Lab of the University of Notre Dame.
Sparse Feature Representations for Presentation Attack Detection in Iris Recognition
In this project, we propose to explore using sparse representations of signals to develop iris image features that would generalize well on unknown attack types. We believe that sparse representations will serve as good features in one-class classification problems (a.k.a. anomaly detection or open-set recognition), when only a positive class (here: live irises) is well populated by samples and the negative class (here: artificial objects) is only modestly represented by training examples, or such training examples do not exist.
LUKSIC ND-UC, in collaboration with the University of Notre Dame.
Video Be On the Look Out
Person re-identification (ReID) is a popular topic of research. Almost all existing ReID approaches employ local and global body features (e.g., clothing color and pattern, body symmetry, etc.). These `body ReID’ methods implicitly assume that facial resolution is too low to aid the ReID process. This project explores and shows that faces, even when captured in low-resolution environments, may contain unique and stable features for ReID. We contribute a new facial ReID dataset collected from a real surveillance network in a municipal rapid transit system. It is a challenging ReID dataset, as it includes intentional changes in persons’ appearances over time. We conduct multiple experiments on this dataset, exploiting deep neural networks combined with metric learning in an end-to-end fashion.
In collaboration with the Computer Vision Research Lab of the University of Notre Dame.
Student Attendance System
Manual attendance sheet management is laborious in crowded classrooms. In this project, we develop a general methodology for an automated student attendance system that can be used in crowded classrooms. The session images are taken by a smartphone camera.
In collaboration with BiometryPass Co.
X-ray testing
Recognition of Prohibited Objects in Baggage Inspection using 3X Strategy
In this project, we propose a unified baggage inspection strategy that we call the `3X-strategy’ that uses a combination of three research areas: X1 (energies: mono-energy, dual-energy, and multi-energy), X2 (views: single view, multiple views, and computed tomography) and X3 (algorithms: of low, medium and high complexity) that can be used in detection processes. We believe that for many threat objects, there can be a suitable combination of them. Our research focuses on how robust X-ray testing can be performed with some degree of generality. Our general goal is to contribute to efforts to improve the effectiveness of baggage inspection by using an ad-hoc 3X combination.
FONDECYT, in collaboration with the University of Notre Dame and Universidad de Atacama.
Andean Geothermal Center of Excellence
At the Andean Geothermal Center of Excellence (CEGA), we work to generate and improve geothermal knowledge in Chile. CEGA is a Fondap-Conicyt project that began operations during the first half of 2011. CEGA comprises a team of researchers from the Faculty of Physical and Mathematical Sciences at the University of Chile, with scientists from the Pontifical Catholic University of Chile and other international institutions. Its five main research fields are Heat Sources, Fluid and Isotopic Geochemistry, Heat-Fluid-Rock Interaction, Structural Controls and Geophysics, Reservoir Architecture, and Modelling. In this project, we use computer vision algorithms to characterize rock properties using 2D and 3D imaging.
FONDAP, in collaboration with the Structural Engineering and Geotechnical Department of the Universidad Catolica de Chile and the Geology Department of the Universidad de Chile.
Intelligent Baggage Inspection System using X-ray Images and Deep Learning
X-ray screening systems have been used to safeguard environments where access control is paramount. Security checkpoints have been placed at the entrances to many public places to detect prohibited items such as handguns and explosives. This project attempts to contribute to object recognition in X-ray testing by evaluating different computer vision strategies based on deep learning proposed in the last few years. We strongly believe that designing an automated aid for the human inspection task is possible using these computer vision algorithms.
UTS KTP Visiting Fellow, in collaboration with the University of Technology Sydney.
Object Segmentation and Recognition in Videos
Traditional object segmentation and recognition methods are built on handcrafted features and shallow trainable architectures. Their performance easily stagnates by constructing complex ensembles that combine multiple low-level image features with high-level context from object detectors and scene classifiers. With the rapid development of deep learning, more powerful tools that can learn semantic, high-level, and deeper features are introduced to address the problems existing in traditional architectures. This project will explore these techniques in online applications such as X-ray screening systems.
MoU NII-UC, in collaboration with the National Institute of Informatics, Japan.
Biology and Medicine
Skin lesion recognition
Skin cancer is a highly relevant health problem worldwide. The World Health Organization (WHO) reports that one-third of the diagnosed cancers are skin cancers. Early skin cancer detection significantly increases patients’ prognosis. In many cases, however, the absence of clinical devices and qualified experts makes this task difficult. In this project, advanced deep-learning techniques have been proposed to recognize skin cancer automatically, showing promising results. This work evaluates different deep-learning approaches on the well-known HAM10000 and our Chilean datasets.
iHealth, Millenium Institute, ANID, Chile.
Characterization of Spinal Cord Damage Based on Automatic Video Analysis of Frog Swimming
Frogs are widely used to study many aspects of modern biology. Their central nervous system is particularly interesting, as in certain stages of metamorphosis, they can regenerate their spinal cord after injury and recover swimming. Therefore, research into automatic gait analysis is important to advance the understanding of these regenerative mechanisms. In this project, we develop a method based on computer vision that automatically and quantitatively establishes the degree of limb movement that can be used to evaluate regeneration performance.
FONDEF, in collaboration with the Center for Aging and Regeneration of the Catholic University of Chile.