Invited Speakers

Hong Zhang is a Professor of Computer Science at Georgia Southern University (USA) and served as the department chair of computer science and information technology for ten years. He received his PhD and MA in Mathematics, MSEE in Electrical Engineering from University of Pittsburgh (USA), and his BS in Computer Science from Fudan University (China). Dr. Zhang has extensive experiences in academia and industry. He has published in a wide range of areas of engineering, computer science, mathematics, and medicine. His current research interests include image processing, machine learning and their medical applications.

Speech Title: Invariant Kernels and Harmonic Analysis

Abstract: Kernels as similarity measures are key components of machine learning algorithms such as Support Vector Machine and Gaussian Process. Invariant kernels are an effective way to incorporate prior knowledge in applications with the invariance property. Harmonic analysis provides a natural mathematical tool in characterization of invariant kernels. On locally compact Abelian groups, Fourier transforms can be defined as an extension of classical Fourier series and Fourier transform. Bochner theorem on functions of positive type is closely related to the invariant kernels defined on the groups. Using this approach, a complete characterization of invariant kernels on the Abelian groups can be obtained. For example, invariant kernels on a circle or a multi-dimensional torus can be explicitly characterized. For non-Abelian groups, even though the kernels cannot be defined with the group operations, harmonic analysis is still relevant in the study of invariant kernels under the group actions. The transitivity of the group actions can provide useful information on the kernel structures. As an example, a characterization of the rotation invariant kernels on spheres is derived. Through the transitivity, it is shown that such a kernel is a function of the dot product of the input vectors alone and the function can be expanded as a series of Gegenbauer polynomials with non-negative coefficients.

Lyudmila Mihaylova, PhD is a Professor of Signal Processing and Control in the Department of Automatic Control and Systems Engineering at the University of Sheffield, Sheffield, United Kingdom. Her research interests are in the areas of autonomous systems with applications to cities, autonomous and assisted living systems. She has expertise in the areas of machine learning, intelligent sensing and sensor data fusion. She won the Tammy Blair best award from the International Conference of Information Fusion 2017, best paper awards from the IEEE DESSERT’2019, 17th IEEE SPA’2013 Conference and IEEE Sensor Data Fusion Workshop, 2013 and others. Prof. Mihaylova is on the Board of Directors of the International Society of Information Fusion (ISIF) and was the ISIF President in the period 2016–2018. She has given a number of talks and tutorials, a including an invited talk in Harbin (Intelligent Navigation and Advanced Information Technology Workshop’2020), plenary talk in Cairo (JIC Smart Cities’2019), NATO SET- 262 AI 2018 (Hungary), Fusion 2017 (Xi’an, China), plenary talks for the IEEE Sensor Data Fusion 2015 (Germany), invited talks at IPAMI Traffic Workshop 2016 (USA) and many others. She was the general vice-chair for the International Conference on Information Fusion 2018 (Cambridge, UK), of the IET Data Fusion & Target Tracking 2014 and 2012 Conferences, publications chair for ICASSP 2019 (Brighton, UK), program chair of Fusion 2020 and others.

Speech Title: Machine Learning Methods for Autonomous Image and Video Analytics

Abstract: Autonomous image and video analytics faces a number of challenges due to the huge volumes of data that sensors provide, changeable environmental conditions and other factors. This talk will discuss current trends in the area of machine learning for image and video analytics in autonomous systems. A key question is how we could develop trustworthy and resilient solutions that are able to work reliably under different conditions? How could we quantify the impact of uncertainties on the developed solutions? Automated detection and behaviour analysis is an important case study which necessitates unsupervised learning algorithms. Recent results for automated video analytics will be presented with Dirichlet process models, deep learning and other methods. Their pros and cons will be discussed.

Chang-Tsun Li is Professor of Cyber Security at Deakin University, Australia. He has had over 20 years research experience in multimedia forensics and security, biometrics, machine learning, computer vision, image processing, pattern recognition, bioinformatics and content-based image retrieval. The outcomes of his research have been translated into award-winning commercial products protected by a series of international patents and have been used by a number of law enforcement agencies, national security institutions and companies around the world, including INTERPOL (Lyon, France), UK Home Office, Metropolitan Police Service (UK), Sussex Police Service (UK), Guildford Crown Court (UK), and US Department of Homeland Security. He publishes actively in prestigious journals such as IEEE T-PAMI, IEEE T-IP, and IEEE -TIFS. He and his co-author are the winner of the Best Paper Award of 2018 IEEE AVSS. In addition to his active contribution to the advancement of his field of research through publication, Chang-Tsun Li is also enthusiastically serving the international cyber security community. He is currently Chair of Computational Forensics Technical Committee (TC-6) of the International Association of Pattern Recognition (IAPR). In the past decade, Chang-Tsun has been active in facilitating the cross-fertilisation of multimedia forensics and biometrics through his leadership in a number of multinational projects and international events.

Speech Title: Imaging Device Fingerprinting

Abstract: Similar to people identification through human fingerprint analysis, multimedia forensics and security assurance through imaging device fingerprint analysis have attracted much attention amongst scientists, practitioners and law enforcement agencies around the world in the past decade. Imaging device information, such as device models and serial numbers, stored in the EXIF are useful for identifying the devices responsible for the creation of the images and videos in question. However, stored separately from the content, the metadata in the EXIF can be removed and manipulated at ease. Device fingerprints deposited in the content by the imaging devices provide a more reliable alternative to aid forensic investigations and multimedia security assurance. Various hardware or software components of the imaging devices leave model or device-specific artifacts in the content in the image acquisition process. These model or device specific artifacts, if properly extracted, can be used as device fingerprints to identify the source devices. This talk will start with an introduction to various types of hand-crafted device fingerprints. Their applications and limitations will then be discussed. Finally, recent data-drive approaches for automatic extraction of imaging device fingerprints will be presented.

Linlin Shen is currently the "Pengcheng Scholar" Distinguished Professor at School of Computer Science and Software Engineering, Shenzhen University, Shenzhen, China. He is also an Honorary professor at School of Computer Science, University of Nottingham, UK. He serves as the director of Computer Vision Institute, AI Research Center for Medical Image Analysis and Diagnosis and China-UK joint research lab for visual information processing. He also serves as the Co-Editor-in-Chief of the IET journal of Cognitive Computation and Systems. He received the Ph.D. degree from the University of Nottingham, Nottingham, UK. He was a Research Fellow with the University of Nottingham, working on MRI brain image processing. His research interests include deep learning, facial analysis and medical image processing. Prof. Shen is listed as the Most Cited Chinese Researchers by Elsevier. He received the Most Cited Paper Award from the journal of Image and Vision Computing. His cell classification algorithms were the winners of the International Contest on Pattern Recognition Techniques for Indirect Immunofluorescence Images held by ICIP 2013 and ICPR 2016.

Speech Title: Know More About You from Your Face

Abstract: As an important personal trait, face has been widely used in real applications for personal authentication. Recently, researchers have also started to look at the possibility of predicting the person’s emotions, depression and personality from his/her facial dynamics, which are very useful for individual understanding and human-robot interactions. In this talk, I will mainly introduce our recent works using facial dynamics to predict the depression and personality of an individual. As the length of video captured for faces during a conversation is different, we firstly extract the temporal dynamics from a number of AUs and them use Fourier transform to get spectral signal with fixed length. The spectral signal is then input to a MLP for depression regression. For personality recognition, our primary work design a self-supervised frame ranking task to train the encoder, then use person-specific cross-layer links to regress the Conscientiousness, dependability, volition, Agreeableness and Openness. Based on this, we further use NAS to search for person specific network to regress the facial reactions. The network architectures are then modeled as graphs for personality recognition. As the Person-specific cognitive processes are modeled, our approach achieves much better performance than state of the art.

Prof. Jiande Sun, received his BS and PhD from Shandong University in 2000 and 2009 respectively. He works with Shandong Normal University at present and serves as the associate dean of the School of Information Science and Engineering. His research interest is on multimedia processing, analysis, understanding and their applications on security, retrieval, interaction, education, etc. Up to now, Prof. Jiande Sun's group have more than 100 papers published in international journals and conference, and over 20 authorized patents for invention.

Speech Title: Cross-Modal Retrieval with Semi- and Multiple Supervision

Abstract: In recent years cross-modal retrieval has attracted widespread attention in both industry and academia areas. Different from the traditional single-modal information retrieval technology, cross-modal retrieval usually uses query data from arbitrary modalities to retrieve semantically relevant instances from other modalities. However, different modalities usually exist in different feature spaces and it is difficult to correlate the low-level features and the high-level semantics. Thus, this leads to the basic challenges in cross-modal retrieval: how to align the low-level "heterogeneous representation" and bridge high-level "semantic gap". This presentation focuses on the multi-class joint subspace learning, semi- and multiple supervised hashing for cross-modal retrieval and the application in fake news detection.

Sabah A. Jassim is a Professor of Mathematics and Computations at the school of Computing, University of Buckingham, UK. He graduated from Baghdad University (BSc & MSc in Mathematics) and holds a PhD in Mathematics from University of Swansea.

Research interest includes Group Actions on Riemann Surfaces, Computational Geometry & Topology, Topological Image Analysis for tamper detection and tumour classification, Biometrics Recognition (Face & Gait), Privacy and Security of multimedia and Biometric templates, and Machine learning for tumour diagnostics. Have more than 150 research papers in refereed journals/conferences, and supervised over 28 PhD theses in Mathematics and Computing. Participated in EU projects (SecurePhone & BroadWan), and innovate-UK funded KTP industry projects, and collaborates with several EU institutes. Is a joint chair of an SPIE conference held annually in US. Held academic posts at several universities (Swansea, De Montfort, City University-London, Fachhochschule Wedel–Germany, and Basrah & Sulaimani Universities in Iraq).

Speech Title: Topological Image Analysis Tools for Detection of Imperceptible Artefacts/Degradation

Abstract: Imperceptible image artefacts/degradation may be due to imperfect capturing device/process, unsuitable environmental recording conditions, malicious tampering, or the presence of abnormal objects in scanned terrains/tissues. We are concerned with the last two cases that challenge the effectiveness of machine learning algorithms for classifying/recognising objects of interest in images. Persistent Homology (PH) is a well-established tool of a recently emerging paradigm of Topological Data Analysis (TDA) and is used for modelling shapes of point clouds. The PH tool investigates persistency of topological invariants of simplicial complexes constructed on these points according to proximity/similarity measures. We shall present and demonstrate significant performance of automatic PH-based algorithms developed for detecting morphing face image attacks, and hint at the potentials of using this approach for medical diagnostic predictions via analysis of scan images of tissues/organs. PH-based image analysis differs from conventional texture image analysis in that it doesn’t only depend on detecting changes to certain texture features or their statistics, but places additional emphasis on the spatial distribution of such changes. Morphing is used to transform one face image (source) to look like another face image (target) thereby seriously undermining the security of Electronic Machine-Readable Travel Document (eMRTD) in Border Control systems.

Binjie Qin received his M.Sc. and Ph.D. degrees from the Nanjing University of Science and Technology, Nanjing, and Shanghai Jiao Tong University, Shanghai, China, in 1999 and 2002, respectively. He was a Lecturer and an Associate Professor with the School of Life Sciences and Biotechnology, Shanghai Jiao Tong University. From 2012 to 2013, He was a Visiting Professor with the Department of Computer Science, University College London, U.K. He is currently an Associate Professor with the School of Biomedical Engineering, Shanghai Jiao Tong University. His current research interests include biomedical imaging, image processing, machine learning, computer vision, and biomedical instrumentation. He has served a reviewer for nearly thirty journals including IEEE TIP, IEEE TNNLS, MIA, PR, IEEE TC, IEEE Access, IEEE TIM, IEEE TUFFC, IEEE SPL, PMB. His group has developed several imaging systems for image guided ENT surgery and fluorescence imaging, which have preclinical and clinical applications in some laboratories and hospitals. Prof. Qin and his group have published more than 60 papers with over 30 papers in SCI/EI indexed journals including IEEE Transactions on Image Processing, Pattern Recognition, Neural Networks, Applied Soft Computing, IEEE Transactions on Instrumentation and Measurement, Physics in Medicine and Biology, IEEE Signal Processing Letters.

Speech Title: Machine Learning for Vessel Extraction in X-ray Coronary Angiography

Abstract: This presentation introduces our model-based and date-driven algorithms within machine learning framework for vessel extraction in X-ray coronary angiography (XCA). First, a motion coherency regularized robust PCA (RPCA) was proposed using low-rank plus sparse decomposition modeling to accurately extract contrast-filled vessel candidates via a statistically structured RPCA, which then subsequently refined vessel extraction via total variation regularization on the trajectory decomposition of the candidate foreground vessels. Second, an encoder-decoder-based sequential vessel segmentation deep network called SVS-net was developed by exploiting several contextual frames of sequential images to segment the vessel masks for the current frame, which was built upon U-net to extract spatiotemporal features in encoder stage and fuse the features in skip connection layers as well as design channel attention mechanism in decoder stage. Third, by exploiting the low-rankness and inter-frame spatiotemporal connectivity in the complex and noisy backgrounds, we recovered the vessel-masked background regions using tensor completion with twist tensor nuclear norm being minimized to complete the background layers. This sequential XCA background completion method accurately extracted foreground vessels' heterogeneous intensities from the noisy XCA data by subtracting the completed background layers from the XCA images. Experimental results showed the superiority of these proposed machine learning-based methods over the state-of-the-art methods. Lastly, we have also applied the concept of deep algorithm unrolling to propose a data-driven neural network-based deep unfolded RPCA tailored to our XCA vessel extraction, which demonstrates improved convergence speed and accuracy with respect to its model-based iterative counterpart.

Dr. Jin is an associate professor in the School of Computer Science and Technology at the University of Science and Technology of China (USTC). His main research interests are big data management, databases, and image retrieval. He received his Ph.D. degree in computer science from the University of Science and Technology of China (USTC) in 2003. After that, he worked as a postdoctoral researcher in the Department of Electronic Engineering & Information Science at USTC. He was a visiting scientist in 2009 at the University of Kaiserslautern, Germany. In 2014, he was a visiting professor at Aalborg University, Denmark.

He is currently serving as an associate editor of the International Journal of Semantic Web and Information Systems and the International Journal of Knowledge Society Research. He also served as the TPC Chair of HardBD’14 and HardBD’13, and the Demo Chair of WAIM’13 and NDBC’12. He has published more than 80 papers in peer-reviewed journals and conferences, including TKDE, TPDS, ICDE, and WWW. Dr. Jin and his students have won two best paper awards (NPC 2014, NDBC 2019), one best student paper award (NDBC 2017), one best paper award runner-up (NDBC 2012), one best demo award (NDBC 2011), and one best poster award (DASFAA 2015).
 

Speech Title: Fast Storage for Big Sensing Data

Abstract: The rapid development of the Internet of Things (IoT) enables people to gather and analyze bigsensing data by deploying various kinds of sensors. As sensing data are usually produced as small files, we have to devise efficient approaches for storing massive small sensor files, which has been a critical issue in IoT data management. However, the small size and high arriving rate of sensor files make the existing Hadoop Distributed File System (HDFS) inefficient to support high write throughput and search performance on massive small sensor files. In this work, we propose the first read/write-optimized solution for storing massive small sensor files on HDFS. In particular, we exploit the distributed caching and parallel file merging policy to improve the write throughput on HDFS. With thismechanism, massive small sensor files are cached by a Distributed Memory File System (DMFS) on top of HDFS, which are furthered merged and flushed to HDFS in a parallel way. We demonstrate that this mechanism is able to provide higher write throughput than HDFS and the existing centralized-cache-based solution. In addition, in order to improve the search performance of object-based queries over massive sensor files, we present the concepts of Sensor Similarity and Sensor Dependence, and further propose a Sensor-Dependence Graph (SDG) to model sensor dependence. Then, we present a SDG-based algorithm to efficiently cluster sensors. Files from sensors in the same cluster are merged into the same big files so as to reduce the file scans and therefore improve the search performance of object-based queries. We conduct extensive experiments to evaluate the performance of our proposal. The results suggest the efficiency of our proposal in search performance, disk write throughput, and memory write throughput.

Dr. Justin Dauwels is an Associate Professor at the TU Delft (Circuits and Systems group). He was an Associate Professor of the School of EEE at the Nanyang Technological University (NTU) in Singapore till the end of 2020. His research interests are in data analytics with applications to intelligent transportation systems, autonomous systems, and analysis of human behaviour and physiology. He obtained his PhD degree in electrical engineering at the Swiss Polytechnical Institute of Technology (ETH) in Zurich. He was a postdoctoral fellow at the RIKEN Brain Science Institute and a research scientist at the Massachusetts Institute of Technology.

He serves as Associate Editor of the IEEE Transactions on Signal Processing, and of the Elsevier journal Signal Processing, and as organizer of several IEEE conferences and special sessions. He is also Elected Member of the IEEE Signal Processing Theory and Methods and IEEE Biomedical Signal Processing TC.

Speech Title: AI for Applications in Psychiatry

Abstract: Many tasks in medicine still involve substantial manual work. In many cases there is strong potential for intelligent automation by A.I., leading possibly to a reduction in costs and man-hours, while increasing the quality of clinical service. In this talk, we will consider applications of A.I. in the domain of psychiatry.
Specifically, we will give an overview of our research towards automated behavioral analysis for assessing the negative symptoms of mentally ill patients.

Qian Huang received the B. Sc. degree in computer science from Nanjing University, China, in 2003, and the Ph. D. degree in computer science from the Institute of Computing Technology, Chinese Academy of Sciences, in 2010. From 2010 to 2012, he was a deputy technical manager of Mediatek (Beijing) Incorporation, Beijing, China. Since Dec. 2012, he is with Hohai University, Nanjing, China, where he serves as the dean of Computer Science & Technology Department.

His research interests include media computing, intelligent education, and intelligent healthcare. He is a member of the Chinese Association for Artificial Intelligence Technical Committee on Deep Learning, a member of the China Computer Federation Technical Committee on Multimedia Technology, and a member of the China Society of Image and Graphics Technical Committee on Multimedia. He holds 6 Chinese patents and 2 US patents, and has published 6 technical books as well as more than 40 technical articles.

Speech Title: Airborne Platform based Ship Identification

Abstract: The rapid economic development continues to drive the waterway transportation industry forward. The huge number of ships and the busy waterway transportation network increase the difficulty of shipping supervision. A smart ship identification system based on an airborne platform is proposed for this purpose. We first build a large-scale ship target recognition dataset called HHU-Ships collected by the near-Earth airborne platform to train and test the ship target detection model. Through the dataset HHU-Ships to compare the advantages and disadvantages of different target detection algorithms, the general target detection model YOLOv3 was finally selected as the basic architecture of the ship target detection model. After obtaining the image of the target ship, we need to recognize the ship's identity. Ship face recognition technology and ship name recognition technology are jointly used to identify ships. In ship face recognition, the SFDS ship face dataset is established to provide data support for research. A mismatch elimination method named GWLMR based on a grid structure and a weighting strategy was proposed to solve the problem of many mismatches caused by factors such as image blur, repeated texture, and scale or viewpoint changes in ship face recognition. In ship name recognition, we constructed a diversified ship name dataset based on the information of ships in China's inland rivers. In order to identify small-scale ship name areas and effectively distinguish between closely distributed multi-line ship names, an attention mechanism-based ship name detection algorithm called SFPENet is proposed. The algorithm introduces an attention mechanism to adaptively fuse the features of each layer of the feature pyramid. In addition, an algorithm called SCONet is proposed for identifying similar ship names based on correction networks. This algorithm improves the positioning network in the spatial transformation network and optimizes the generation strategy of control points. The ship name matching algorithm based on the Chinese ship name database is proposed as a optimization measure after recognition processing to improve the overall recognition accuracy. Experiments have proved that our method can effectively locate and identify ships.

Mr. Hidekazu Tsujiai was received the B.Sc. in Science (Applied Mathematic) from Konan University, Kobe in 1983. I was received M.A. in Integrated Arts and Sciences (Mathematics and Information Sciences) from Osaka Prefecture University, Osaka in 1986. I was received D. Eng. in Engineering from Osaka Prefecture University, Osaka in 2000. I was Research associate in Dept. of Electronic, School of Science and Technology, Kindai University from 1987 to 1993. I was Assistant professor in Dept. of Electronic System and Information Engineering, School of Biology-Oriented Science and Technology, Kindai University, Wakayama, from 1993 to 2005. I am Associate professor in Faculty of Art and Design, University of Toyama, since 2005. I won the Japan Society for Graphic Science Award in 2015. I won the Hagura Prize of the Art Technologies Expression Association in 2018. I was Senior Member of the IEEE. My research is a fusion of art and science.

Speech Title: Making Fulldome Images Using Sphere Camera and Fulldome Picture on Bowl

Abstract: This presentation is an improvement on the paper “Making Fulldome Images using Camera with Fisheye Lens and Fulldome Picture on Bowl Workshop for Children at Planetarium”. In the previous paper, the starting point was the lack of a drawing workshop for children in the planetarium. With a simple structure that does not require internal lighting when shooting inside the ball, the picture drawn inside the announced styrene ball could be taken with a fisheye lens. Even children could draw inside the ball, and the images taken with the fisheye lens could be projected on the planetarium. In this study, the internal structure was reduced by changing from a camera with a fisheye lens to a spherical camera. As the camera be-came smaller, the shooting range changed from a hemisphere to a sphere. Even without a planetarium, the system could be experienced anywhere with VR glasses.

Daisuke Miyazaki received the BS degree in science from the University of Tokyo in 2000, the MS degree in information science and technology from the University of Tokyo in 2002, and the PhD degree in information science and technology from the University of Tokyo in 2005. He is an associate professor at Hiroshima City University, Japan. After working at the University of Tokyo for two and a half years, the Microsoft Research Asia for a half year, and the University of Tokyo again for a half year, he joined Hiroshima City University. He is an Editorial Board of International Journal of Computer Vision & Signal Processing. He received the best paper award from VSMM2000 and FCV2020. His research interests include computer vision. He is a member of ACM and IEEE, and a senior member of IEEE.

Speech Title: Shape from Polarization

Abstract: Polarization is a well-known physical phenomenon, and there are variety of researches proposed until now. Polarization has a wide area of application not only in physics which for example analyze the material structure but also in entertainment which for example uses stereo projecting screen. Researches about polarization are performed mainly in the field of optics, but recently, they are also performed in other fields. This talk briefly explains the basic theory of polarization. In addition, this talk introduces some researches about shape-from-polarization which are performed in the field of computer vision.