Dr. Iman Yi Liao obtained her BSc, MSc, PhD all from School of Computer Science and Engineering, Northwestern Polytechnical University, Xi'an, China, in 1996, 1999, and 2006 respectively. She was a visiting researcher in Chinese University of Hong Kong from July 2001 to Dec 2001, and in CVSSP, University of Surrey, UK from Dec 2003 to Sep 2004, respectively. She was a Post-Doctorate research fellow at the School of Computer Sciences, Universiti Sains Malaysia in 2008 and subsequently a Senior Lecturer from 2009 to 2012. She joined the School of Computer Science, University of Nottingham Malaysia Campus, as an Associate Professor in December 2012.
I am interested in innovating and applying Machine Learning techniques to various Computer Vision problems, including 3D reconstruction from single 2D images, 3D point cloud registration and classification, landmark detection of objects in 2D/3D images, semantic segmentation of 2D/3D images, object detection and classification in 2D images. My recent research focuses on industrial applications where data are often collected with small sample size.
Theoretical tools I have investigated include but not limited to variational methods, optimization theory and algorithms, regularization methods, relaxation algorithms, fractal analysis, multi-scale analysis (e.g., wavelets), Markov random fields, B-Splines, Differential Geometry, Principal Component Analysis, and some typical Machine Learning techniques as well as Deep Convolutional Neural Networks.
The current modules she is teaching include
- Mathematics for Computer Scientists
- Computer Vision
- Machine Learning
- Advance Algorithms and Data Structure
Other modules she has taught/ was involved in include
- Computer Graphics
- Software Engineering
My current research includes both theoretical and application works.
In the theoretical aspect, currently I am focusing on Small Sample Size Problem in High Dimensional Data Modeling, a challenging problem that has recently drawn attention of researchers from various fields such as genomics, Image and Video Analysis, Chemometrics, Economics, and Humanities. As a common problem in the above mentioned areas, where the classic statistical theories have difficulty (as they work based on assumption of large number of examples as compared to data dimensionality), it requires exploring new theories and techniques in statistical and machine learning, which can be hopefully applied to solve related problems in all those aforementioned areas.
As for the application of image processing and computer vision, I have worked with medical doctors and forensic experts at Hospital Kuala Lumpur on Post Mortem Computerised Tomography (PMCT) data, to analyse and identify the associated demographics such as gender, age, race, etc, using image processing and machine learning techniques apart from standard methodologies in Forensic Medicine and Sciences. I have also collaborations with Universiti Kebangsaan Malaysia, Monash University Malaysia, and several industrial partners, on various Image Processing and Computer Vision researches/projects.
MAHMOOD HAITHAMI, AMR AHMED, IMAN YI LIAO and HAMID JALAB, 2023. Enhancing polyp segmentation generalizability by minimizing images' total variation In: the IEEE International Symposium on Biomedical Imaging (ISBI) 2023. (In Press.)
MAHMOUD A. AZIM KHATTAB, IMAN YI LIAO, EAN HIN OOI and SIANG YEW CHONG, 2022. Compound W-Net with fully accumulative residual connections for liver segmentation using CT images Computational and Mathematical Methods in Medicine: Machine Learning and Artificial Intelligence Methods in Computer Vision and Visualization for Healthcare. IMAN YI LIAO, ERIC SAVERO HERMAWAN and MUNIR ZAMAN, 2022. Body Landmark Detection with an Extremely Small Dataset Using Transfer Learning Pattern Analysis and Applications.
MAHMOOD HAITHAMI, AMR AHMED, IMAN YI LIAO and HAMID JALAB, 2022. Automatic bowel preparation assesment using deep learning In: 2nd International Workshop on Artificial Intelligence for Healthcare Applications (AIHA 2022), International Conference on Pattern Recognition (ICPR2022).
I have conducted research on 3D Terrain Reconstruction from Remote Sensing Images, 3D Face Reconstruction from Single 2D Images, Automatic Landmarks Detection and Placement on 3D Cranio-Facial Data, and some preliminary work on Computer-Aided Anthropology in Forensic Science. I also co-researched on Mobile Robot Localization and Vision-based Techniques.
My doctoral research was on reconstruction of 3D surfaces from single 2D images with fractal prior, falling under the ambit of Shape from Shading (SFS). While conventional methods relied on constructing smoothness constraints to regularise the problem, I took a revolutionary view and introduced a fractal priori that is able to recover rough, as well as smooth surfaces, more accurately.
When dealing with small sample size problem in Machine Learning, I have developed several methods for the problem of 3D face reconstruction from 2D single images, including an adaptive model based on generating artificial samples and a latent variable model based on general human traits such as gender, age, and racial groups.
In domains of very high dimensional sample data, a compact representation of original data is obtained before further processing. We have proposed a method that takes the full advantage of feature extraction and simultaneously relate features to the importance of anatomical regions on skulls for sex determination. We have also provided a full automated process that takes in the skull CT scan images and estimates the sex of the subject. Our work was considered "first of a kind" by peer review and the prototype we developed was invited to the International Conference and Exposition on Inventions by Institutions of Higher Learning (PECIPTA 2017), in Terrengganu, Malaysia, 2017.
My recent research involves solving problems in industry where the data sample size is small. We have designed novel methods leveraging on transfer learning from existing Deep Learning solutions/models to address such issues. In one of the recent case, we have successfully designed a deep convolutional neural network model for detecting body landmarks for customised tailoring services based on transfer-learning from an existing model that was dedicated for fashion landmark detection. The model has been successfully deployed by the partner company.
With my previous research experience in 3D reconstruction and 3D vision data processing, and my current interest on example based modeling, I would hope to explore and solve research problems with applications to Geophysical Science, Forensic Science, Medical Image Analysis/Retrieval and any other related areas that can make full use of vision and image analysis, statistical learning, and pattern recognition techniques.
You are welcome to collaborate with me if you see any possibilities that I can contribute to your research areas.