- Perception, Action and the Information Knot that Ties Them - July 22, 2015
- Deep Learning Image Understanding with Subspace Indexing on Grassmann Manifolds - July 2, 2015
- Quantitative Brain Image Analysis: From Macro to Nano - May 7, 2015
- Bonirob - an autonomous mobile platform for agricultural robotics - March 17, 2015
- Is the Gaussian distribution "Normal"? Signal Processing with Alpha-Stable Distributions - March 13, 2015
- Statistical modelling of spatio-temporal signals: from the Afghan conflict to BOLD-MRI in rats' kidneys - March 5, 2015
- Saliency Based Selection of Gradient Vector Flow Paths for Content Aware Image Resizing - November 25, 2014
- Computer Vision and the Modeling and Reverse Engineering of Biological Systems - October 31, 2014
- Synchronisation and matching in camera networks - March 27, 2014
- Processing and Analysis of Mars Missions Data - December 13, 2013
- Imaging and Image Processing for Plant Phenotyping at Forschungszentrum Jülich - November 7, 2013
- From gigapixel time-lapse and UAVs to food security and adaptive conservation: software and hardware requirements for enabling NextGen ecology and phenomics - October 23, 2013
- Non-contrast MRI: applications in cardiac and peripheral muscle - October 16, 2013
- The ABC of Ecoinformatics - Knowledge representation, Agents, Services, and Programming Languages for Environmental Modelling and Decision Support - May 8, 2013
- Sparse and Redundant Representations: Theory and Applications - March 25, 2013
Perception, Action and the Information Knot that Ties Them
I will describe a notion of Information for the purpose of decision and control tasks, as opposed to data transmission and storage tasks implicit in Communication Theory. It is rooted in ideas of J. J. Gibson, and is specific to classes of tasks and nuisance factors affecting the data formation process. When such nuisances involve scaling and occlusion phenomena, as in most imaging modalities, the “Information Gap” between the maximal invariants and the minimal sufficient statistics can only be closed by exercising control on the sensing process. Thus, senging, control and information are inextricably tied. This has consequences in understanding the so-called “signal-to-symbol barrier” problem, as well as in the analysis and design of active sensing systems. I will show applications in vision-based control, navigation, 3-D reconstruction and rendering, as well as detection, localization, recognition and categorization of objects and scenes in live video.
Speaker: Stefano Soatto, UCLA
July 22, 2015, 14:30, San Francesco − Classroom 2
Deep Learning Image Understanding with Subspace Indexing on Grassmann Manifolds
In large scale visual pattern recognition applications, when the subject set is large the traditional linear models like PCA/LDA/LPP, become inadequate in capturing the non-linearity and local variations of visual appearance manifold. Kernelized non-linear solutions can alleviate the problem to certain degree, but faces a computational complexity challenge of solving an Eigen problems of size n x n for number of training samples n. In this work, we developed a novel solution by applying a data partition on the BIGDATA training set first and obtain a rich set of local data patch models, then the hierarchical structure of this rich set of models are computed with subspace clustering on Grassmanian manifold, via a VQ like algorithm with data partition locality constraint. At query time, a probe image is projected to the data space partition first to obtain the probe model, and the optimal local model is computed by traversing the model hierarchical tree. Simulation results demonstrated the effectiveness of this solution in capturing larger degree of freedom (DoF) of the problem, with good computational efficiency and recognition accuracy, for applications in large subject set face recognition and image tagging.
Speaker: Zhu Li, Samsung Research America
July 2, 2015, 11:00, San Francesco − Classroom 2
Quantitative Brain Image Analysis: From Macro to Nano
Continuous progress achieved in medical imaging technology in the past decades has lead to considerable improvement in patient care, but also increased the need for automated tools to analyze and interpret the growing amount of medical data acquired. In an attempt to answer this need, we carry out collaborative research efforts aimed at the adaptation and development of novel technologies for the analysis of medical images. Accordingly, this talk will summarize our ongoing research efforts focused on the human brain. In the first part of the talk we will explore automated analysis of structural brain MR images for dementia. Then in the second part we will introduce preliminary outputs of our ongoing research activity on the analysis of dendritic spines. We will conclude the talk by some open problems and future directions.
Speaker: Devrim Ünay, Bahçeşehir University
May 7, 2015, 10:00, San Francesco − Sagrestia
Bonirob - an autonomous mobile platform for agricultural robotics
Agriculture and related tasks consist of successively executed and well-defined sub-processes. Thus, they appear as prime examples for applying autonomous / robotic technologies. Such an approach seems to be particularly appropriate for individual plant treatment requirements (e.g. for high value crop), analysis and scouting, and organic farming, and it has the potential to lead to zero-tillage agriculture in specific fields. Bosch, Amazone, and the University of Applied Science Osnabrück have developed the robotic platform BoniRob. It provides a high maneuverability and the generic functions to allow for an autonomous navigation. Furthermore, to increase the usability of BoniRob, a platform / App-concept is foreseen. An “App” is understood as a task specific combination of sensors, actuators, and a function which can easily be exchanged on the multi-purpose platform. The latter navigates on the field along the plant rows, carrying the working app. In this regard several prototypical apps already have been implemented, e.g. vision based weed control apps, soil penetrometer, and sensor systems for field-based scouting.
Speaker: Markus Höferlin, BOSCH
March 17, 2015, 11:00, San Francesco − Classroom 1
Is the Gaussian distribution "Normal"? Signal Processing with Alpha-Stable Distributions
There are solid reasons for the popularity of Gaussian models. They are easy to deal with, lead to linear equations, and they have a strong theoretical justification given by the Central Limit theorem. However, many data, manmade or natural, exhibit characteristics too impulsive or skewed to be successfully accommodated by the Gaussian model. The wide spread power laws in the nature, in internet, in linguistics, biology are very well known. In this talk we will challenge the "Normality" of the Gaussian distribution and will discuss the alpha‐stable distribution family which satisfies the generalised Central Limit Theorem. Alpha‐Stable distributions have received wide interest in the signal processing community and became state of the art models for impulsive noise and internet traffic in the last 20 years since the influential paper of Nikias and Shao in 1993. We will provide the fundamental theory and discuss the rich class of statistics this family enables us to work with including fractional order statistics, log statistics and extreme value statistics. We will present some application areas where alpha‐stable distributions had important success such as internet traffic modelling, SAR imaging, computational biology, astronomy, etc. We will identify open problems which we hope will lead to fruitful discussion on further research on this family of distributions.
Speaker: Ercan Kuruoglu, CNR Pisa
March 13, 2015, 11:00, San Francesco − Classroom 1
Statistical modelling of spatio-temporal signals: from the Afghan conflict to BOLD-MRI in rats' kidneys
Spatio-temporal data is ubiquitous in most application domains, from social sciences to medicine and environmental sciences. In this talk I will describe how we have tackled two different modelling problems from two very different domain applications. In the first part of the talk, I will review dynamical spatio-temporal statistical modelling, and show how it could be applied to model and predict the behaviour of the Afghan conflict from the Wikileaks data set. I will then move to a biomedical application where we hoped to use similar methodologies to describe the dynamics of blood oxygenation in rat kidneys following administration of drugs. Unfortunately, in this case the data was not sufficient to parametrize fully a dynamical model, hence we resorted to an empirical model which was still able to reveal interesting biological behaviours in response to treatments.
Speaker: Guido Sanguinetti, University of Edinburgh
March 5, 2015, 15:00, San Francesco − Classroom 1
Saliency Based Selection of Gradient Vector Flow Paths for Content Aware Image Resizing - November 25, 2014
Saliency Based Selection for Content Aware Image Resizing Content-aware image resizing techniques allow to take into account the visual content of images during the resizing process. The basic idea beyond these algorithms is the removal of vertical and/or horizontal paths of pixels (i.e., seams) containing low salient information. In this talk a method which exploits the Gradient Vector Flow (GVF) of the image to establish the paths to be considered during the resizing will be presented. The relevance of each GVF path is straightforward derived from an energy map related to the magnitude of the GVF associated to the image to be resized. To make more relevant the visual content of the images during the content-aware resizing, we also propose to select the generated GVF paths based on their visual saliency properties. In this way, visually important image regions are better preserved in the final resized image. The proposed technique has been tested, with good performances, both qualitatively and quantitatively, by considering a representative dataset of 1000 images labeled with corresponding salient objects (i.e., ground-truth maps).
Speaker: Sebastiano Battiato, Università degli Studi di Catania
November 25, 2014, 11:00, San Francesco − Classroom 2
Computer Vision and the Modeling and Reverse Engineering of Biological Systems - October 31, 2014
Modeling and reverse engineering of complex biological systems is a major focus of current research in bioengineering, biomedicine and computational biology. The key issue of system identification, i.e., determining the transfer function(s) and feedback function(s) from observations of input, output and state variables, is essentially a problem in reconstruction. Computer vision offers a means to observe and model complex biological systems and processes in a non-contact, non-destructive, minimally invasive manner while also improving system observability. This talk will focus on two applications of computer vision in the field of biomedicine, i.e., reconstructive craniofacial surgery and modeling of the endocardial surface of the human heart. In the case of reconstructive craniofacial surgery, computer vision is shown to offer a tool for presurgical planning and simulation that reduces the time for actual surgery. In the case of endocardial surface modeling, computer vision is shown to offer a minimally invasive means for diagnosis of coronary artery disease.
Speaker: Suchi Bhandarkar, The University of Georgia
October 31, 2014, 10:00, San Francesco − Classroom 2
Synchronisation and matching in camera networks - March 27, 2014
Networks of cameras are composed of nodes with the capability of performing local processing that helps transferring the minimum amount of information for completing network tasks. Two important challenges for camera networks are synchronisation (temporal video alignment) and object matching (inter-camera association and object re-identification). In this seminar I will present a video alignment method based on observing the actions of a set of articulated objects that is applicable to general and unconstrained scenarios in a way that is not feasible with current state-of-the-art approaches. The method uses high-level video analysis (object actions) and does not impose constraints on the relative pose or motion of the cameras, on the structure of the time warping between the videos and on the amount of overlap among the fields of view. Next I will discuss a feature selection method that minimizes the data needed to represent the appearance of an object by learning the most appropriate features for person re-identification. For each feature, the computational cost for extraction and storage are considered together with its performance to select cost-effective good features. This selection allows us to improve the re-identification while reducing the bandwidth for communicating the features over the network. The methods will be illustrated with several examples from on-going and recently completed projects.
Speaker: Andrea Cavallaro, Queen Mary University of London
March 27, 2014, 10:00, San Francesco − Classroom 1
Processing and Analysis of Mars Missions Data - December 13, 2013
Mars, among the planets of our Solar System, is the nearest and the most similar to the Earth. Several scientific missions have been sent to Mars in the last 15 years, and others will be sent in the next years. Main objective of missions to Mars is investigate the red planet in order to understand: i) the geological and climatic evolution of the planet, ii) if Mars could have hosted forms of life in the past, and iii) if Mars could be habitable from humankind in the future. In this seminar, planet Mars will be briefly introduced; then, procedures for achieve, process and analyze remote-sensed data from NASA and ESA Mars missions will be illustrated. Challenges and open problems related to Mars missions data processing and analysis will be emphasized.
Speaker: Andrea Pacifici, International Research School of Planetary Sciences
December 13, 2013, 14:00, Ex Boccherini − Conference Room
Imaging and Image Processing for Plant Phenotyping at Forschungszentrum Jülich - November 7, 2013
The Institute of Bio- and Geosciences, IBG-2: Plant Sciences at Forschungszentrum Jülich investigates plants and their dynamic interaction with the environment. We target at a deep understanding of the connections between the expressed plant structure and behavior, i.e. the phenotype, and the genetic causes, i.e. the genotype. Investigations are done on different spatial and temporal scales, from microscopic cell structures to fields scales, and from seconds to months. This talk presents different imaging modalities and devices along with image processing solutions for plant phenotyping developed at IBG-2. Applications are e.g. root and leaf growth analyses using optical-flow-like parameter estimation schemes, or structural and functional analyses on plant roots using MRI and PET.
Speaker: Hanno Scharr, Forschungszentrum Jülich
November 7, 2013, 10:00, Ex Boccherini − Conference Room
From gigapixel time-lapse and UAVs to food security and adaptive conservation: software and hardware requirements for enabling NextGen ecology and phenomics - October 23, 2013
NextGen ecology and genomics will require much greater collaboration between biologists computer scientists and hardware engineers. Many solved problems in computer science and image analysis are still viewed as great challenges in plant sciences and ecology. Likewise, many emerging hardware technologies show great promise for improving ecology but are still technically complex to implement and so remain of limited use to biologists. I will discuss the hardware and software challenges for developing both high throughput plant imaging systems and NextGen landscape monitoring systems for field applications. In the lab we are developing a high throughput software/hardware pipeline for quantifying growth in up 2000 plants in climate chambers with multi-spectral LED lighting and dynamic environmental conditions. Any aspect of plant growth that can be quantified from time-lapse imagery can be tested for genetic association in our custom Genome Wide Association (GWAS) bioinformatics pipeline. In our field work we are seeking to extend standard environmental monitoring by orders of magnitude through gigapixel imaging systems, microclimate mesh sensor networks, UAVs and other emerging technologies. Enabling these efforts requires combining complex hardware and custom software into user-friendly turn-key systems.
Speaker: Timothy Brown, Australian National University
October 23, 2013, 17:00-18:30, Ex Boccherini − Conference Room
Non-contrast MRI: applications in cardiac and peripheral muscle - October 16, 2013
Many diseases may cause renal insufficiency, including diabetes and hypertension. Patients with renal insufficiency cannot undergo conventional contrast enhanced diagnostic tests, such as CT angiography and first-pass MRI perfusion imaging due to the risk of nephrotoxicity or nephrogenic systemic fibrosis. An alternative imaging method is to apply non-contrast MRI techniques. In this presentation, each of non-contrast MRI methods for imaging myocardial perfusion, myocardial viability, and even skeletal muscle perfusion will be critically reviewed. Studies from us and others will be included to show the promise of these methods for the diagnosis of cardiovascular diseases without sticking any needle.
Speaker: Jie Zheng, Washington University in St. Louis
October 16, 2013, 16:00-17:30, Ex Boccherini − Conference Room
The ABC of Ecoinformatics - Knowledge representation, Agents, Services, and Programming Languages for Environmental Modelling and Decision Support - May 8, 2013
Ecology, environmental engineering and natural resource management comprise complex and challenging domain for demonstrating Computer Science algorithms and IT tools. Environment is a broad, interdisciplinary field, and both natural process modelling and decision support for environmental management require a significant investment from software engineers to understand the domain, while raise new research challenges. Environmental applications are not only data-intense and complex, but also inherit from natural systems problems of uncertainty and scaling, which altogether comprise the key challenges of ecoinformatics research. This introduction provides a gentle introduction to ecoinformatics, i.e. the science of information in ecology and environmental sciences, by presenting key concepts and research challenges. We investigate complex environmental applications of computer science and the enabling role of computer science for investigating nature. Technology focuses on knowledge representation, multi-agent systems, service-oriented software engineering and domain-specific programming languages; and is accompanied with applied research results in several domains, including agriculture, water management, air quality, ecosystem services, and meteorology.
Speaker: Ioannis N. Athanasiadis, Democritus University of Thrace
May 8, 2013, 14:00-15:30, Ex Boccherini − Conference Room
Sparse and Redundant Representations: Theory and Applications - March 25, 2013
In this talk we describe some of the recent advances in sparse and redundant representations of signals. Specific applications are described ranging from the matrix completion problem, to video retrieval and compressive sensing. Such problems are formulated and solved using deterministic and stochastic approaches. Specific examples are shown from image and video processing and comparisons are made with the state-of-the-art algorithms. Open problems and future research directions are discussed.
Speaker: Aggelos K. Katsaggelos, Northwestern University
March 25, 2013, 15.30-17.00, Ex Boccherini - Conference Room
Full list of previous seminars see here: IMT Seminars.