Image Processing Seminar Series
Series organiser : Toby Breckon, lecturer in computer vision and image processing
Forthcoming AMAC seminars
As a guide the series generally runs from Jan. to Feb. each year to fit in with the image processing MSc programme. Other seminars will also appear on the above AMAC seminar schedule.
2012
-
Machine Learning for Computer Vision: Randomised Decision Forests and their Novel Applications - Friday 17th Feb. 2012
This talk begins with an overview of machine learning studies for computer vision. Various visual recognition tasks can be cast as a learning problem of mapping from input to output data with emphasis on generalisation to unseen data. Challenges occur due to high-dimensional input space and data changes by different camera poses, illuminations as well as wide intra-class variations. Among state-of-the-art techniques, we emphasise randomised decision forests, an emerging technique, and tree-structured methods, for efficiency. Following concepts and principles, their applications are demonstrated for challenging novel problems: real-time action recognition, object phenotype recognition using 3D shape priors, and video-based object recognition, all of which were recently carried out at Imperial College London (jointly with University of Cambridge).
Tae-Kyun Kim Imperial College, London
Venue: White Lecture Threatre, Building 52, Cranfield University
-
EFIT-V: Creating faces from numbers - Friday 24th Feb. 2012
A facial composite is a graphical representation of a suspect’s face that is generated from an eyewitness's memory. In the absence of other forensic evidence, a facial composite may constitute the only means of locating a suspect. Until recently all systems for constructing facial composites, such as E-FIT (Electronic Facial Identification Technique), were essentially electronic versions of previous mechanical feature-based systems (e.g. PhotoFIT and Identikit). The effectiveness of the feature-based approach is fundamentally limited by the witnesses’ ability to recall and describe accurate facial features from memory. EFIT-V is a novel approach to composite construction in which the witness is simply required to recognise similarities between computer-generated faces and the suspect’s face. A low-dimensional model from which an approximate likeness to any face can be generated. However, the model variables corresponding to the suspect’s facial appearance are not known in advance and must be determined by an iterative selection using an interactive evolutionary algorithm.
Stuart GibsonSchool of Physicial Sciences, University of Kent
Venue: White Lecture Threatre, Building 52, Cranfield University
2011
-
Sketch based multimedia retrieval - Friday 28th January 2011
Image and video databases are commonly searched using keywords. Yet text can be an inefficient vehicle for describing the visual properties of a scene, or the interactions of objects within it. This talk will present techniques being developed at CVSSP for searching media repositories using visual queries, specifically free-hand sketches.
We first describe a technique for searching video using "storyboard sketches"; free-hand sketches depicting both objects present in a scene, and their movements (using motion cues such as streak-lines and arrows, borrowed from production storyboarding). A probabilistic model of object shape and motion is created from the sketched query, and fitted to video clips to determine their support for the sketch.
The model seeks to aggregate super-pixels within the video clip to form objects that approximately match the sketch's colour distribution, shape and motion parameters. We demonstrate matching over a database of sports footage.
We then outline recent work towards a system for retrieving and localising sketched shapes within photos. This is achieved using a novel "Gradient Field HoG" descriptor within a bag-of-visual-words retrieval framework. We demonstrate a prototype photo montage application driven by our retrieval system, and present a quantitative comparison against state of the art descriptors such as Self-similarity, classic HoG and SIFT.
John Collomosse CVSSP, University of Surrey
Venue: Red Lecture Threatre, Building 52, Cranfield University
-
Computer Vision at the BBC - Friday 4th February 2011
Computer Vision is becoming more ubiquitous throughout the programme making chain, from automated production, to augmented reality through to new possibilities when delivering the content to the end user. A brief overview of some of the applications of Computer Vision in broadcasting with specific examples of its use at the BBC is presented here. New production tools are becoming available in tapeless production with regard to automated metadata production. This metadata can be used to streamline the editing process freeing up resources for artistic decisions. Metadata is important throughout the production chain, and new uses for this metadata are being researched in the world of archiving, specifically for the case of programme classification. As well as for metadata production, Computer Vision is becoming increasingly important in the production of the content itself, through the use of augmented reality, both at the point of content capture and in post production. Finally, opportunities for new ways of content delivery are being explored as exemplified by "The Cut Up", an interactive art piece displayed at the anti-design festival.
Jonathan Mckinnell BBC R&D
Venue: Red Lecture Threatre, Building 52, Cranfield University
-
The Power of Processing - Friday 11th February 2011
Waterfall are involved in state-of-the-art military processing, simulation and modelling technology to wider markets, and provide cost-effective solutions to your imaging system requirements. In addition to our “off-the-shelf” hardware and software systems, the majority of our systems are supplied as bespoke, one-off units tailored to specific imaging requirements. We live in a 3D world, but most imaging technology gives 2D output because of the challenges involved. However, viewing in 3D is more intuitive for humans and allows more information to be conveyed about the scene. Waterfall Solutions is pioneering 3D imaging techniques to develop cost-effective systems that can be used with a wide range of cameras, from every-day camcorders to high quality thermal imagers in additon to a wide range of other research and development activity in the imaging arena.
Mark Bernhardt Waterfall Solutions
Venue: Red Lecture Threatre, Building 52, Cranfield University
-
Multimedia information retrieval: What has a falcon got to do with baseball? - Friday 18th February 2011
Prof. Stefan Rueger Knowledge Media Institute, The Open University
Venue: White Lecture Threatre, Building 52, Cranfield University
-
Space Filling Curves and Applications - Monday 21st November 2011
A continuous curve mapping an unit interval into a unit hypercube is called space-filling curve or Peano curve. There are several applications using this kind of curves such as image processing, database retrieval, etc. In this seminar, the definition and the properties of the space filling curve, and the applications are introduced. In 1890, G. Peano has found a curve which passes through all points in a space and has proved its existence – today we see its application in image processing and pattern matching.
Prof. Sei Kamata Waseda University, Japan
Venue: White Lecture Threatre, Building 52, Cranfield University
-
Brain-inspired VLSI Vision Systems for Robot and Vehicle Applications - Monday 21st November 2011
Toward applications to robot or intelligent-car vision, real-time natural-image processing systems inspired by various brain-like models and functions have been constructed. The models are coupled MRF models including resistive-fuse networks for coarse image-region segmentation, Gabor wavelet transformation for feature extraction, a directional state-propagation model for subjective contour generation, and elastic-graph matching for flexible image recognition. Various VLSI chips implementing these models have been designed based on a merged analog-digital circuit architecture.
Prof. Takashi Moire Kyushu Institute of Technology, Japan
Venue: White Lecture Threatre, Building 52, Cranfield University
2010
-
Multimedia information retrieval: What has a falcon got to do with baseball? - Friday 12th February 2010
We used to search authors or titles in library cards in order to locate relevant books; now we can issue keyword searches within the full text of whole book repositories to identify the authors, titles and locations of relevant books. What about the corresponding challenge of finding multimedia by fragments, examples and excerpts? Rather than asking for a music piece by artist and title, can we hum its tune to find it? Can doctors submit scans of a patient to identify medically similar images of diagnosed cases in a database? Can your mobile phone take a picture of a statue and tell you about its artist and significance via a service that it sends this picture to?
Prof. Stefan Rueger Knowledge Media Institute, The Open University
Venue: White Lecture Threatre, Building 52, Cranfield University
2009
-
Framestore and The Golden Compass: Tools and Techniques of VFX - 30th January 2009
An inside look at the how Framestore's Oscar winning visual effects work on The Golden Compass, using digital techniques to create the polar bears and arctic environments.
Andy Lomas Head of CG, Framestore
Venue: Gold Lecture Threatre, Building 52, Cranfield University
-
Threat Detection for Autonomous Reconnaissance Robots - 13th February 2009
A talk about our project to develop recognition software for autonomous reconnaissance robots. The project consisted of development of a swarm of autonomous flying vehicles navigated via wireless link and a GPS system. The UAVs acquire images which are then processed in the ground station by threat recognition software. Our recognition system is based on a vocabulary forest of local features. Large numbers of features are extracted from image examples and are represented by many vocabulary trees. Features from a query image are matched to the trees and vote for object categories and their locations. Large number of trees make the process efficient and robust. The system is capable of simultaneous categorization and localization of objects. The approach obtains excellent performance on large test sets. We demonstrate the robustness of our method to appearance variations, scale change, asymmetric objects, background clutter and occlusion. The system has been extended to human action recognition by incorporating motion information into local features. The performance has been evaluated on 17 challenging real action categories from Olympic Games.
Krystian Mikolajczyk University of Surrey, UK
Venue: Green Lecture Threatre, Building 52, Cranfield University
-
Real time fused imaging, target detection, and tracking. - 18th February 2009
Real time fusion of imagery from two or more complementary sensors offers significant operational benefits for both operator in the loop and automatic processing systems. The presentation reviews the benefits of intelligently applying image fusion in terms of a performance and operational trade space assessment. The assessment characterises the opportunities for intelligently applying fusion in terms of performance, processing integration and hardware implementation. A number of task based examples are used to illustrate the approach taken and report on the quantitative and qualitative assessment of the processing architecture.
James Sadler Waterfall Solutions Ltd.
Venue: White Lecture Threatre, Building 52, Cranfield University
-
CCTV analytics: some recent advances - 27th February 2009
The talk will describe current progress in the growing area of video analytics: the automatic capability of detecting situations of interest especially in the context of monitoring human activity. Capabilities being developed in a number of research labs will be illustrated as well as what is now becoming commercially available. This is a critical look at this emerging technology, highlighting some advances and limitations.
Sergio Velastin Kingston University
Venue: White Lecture Threatre, Building 52, Cranfield University
2008
-
Software Development for Ground Robotics - 25th January 2008
A first hand overview of software roles in front line ground robotics development and the challenges faced in the real world systems deployment with an insight into current technologies in use and sought in this area.
Stephen Bridges, Systems Engineer (Dismounted & Force Protection, QinetiQ)
Venue: Gold Lecture Threatre, Building 52, Cranfield University
-
Camera sensors: the pros and cons of CCD and CMOS - 1st February 2008
Digital cameras use one of two sensor technologies - CCD (Charge- Coupled Device) or CMOS (Complementary Metal Oxide Semiconductor). The received wisdom is that CCD sensors give better images but CMOS sensors are cheaper. This seminar looks at the state of the art to show that this is an over-simplified view. Although CCD sensors lead the market for most industrial cameras, CMOS technology is improving all the time. The two types of sensors are compared and a new generation of CMOS sensor is presented that challenges CCD for quality.
Jon Vickers, Technical Manager (Firstsight Vision)
Venue: Blue Lecture Theatre, Building 52, Cranfield University
-
Advanced Video Processing - 8th February 2008
Building upon a long history of computer vision processing for motion capture, 2d3 developed the Emmy award winning automatic camera tracker Boujou and have recently entered the defence industry exploiting the same underlying technology. This talk will cover a brief history of the company followed by details of current and future projects concentrating on unmanned aerial vehicle (UAV) applications.
James Srinivasan, Development Engineer (2d3)
Venue: Blue Lecture Theatre, Building 52, Cranfield University
-
Medical Imaging - Seeing Inside the Human Body - 13th February 2008
Understanding how the human body is built, how it functions and how it goes wrong has been greatly assisted by the imaging technology to see beneath the skin without the need to cut it open. Medical imaging relies not only on the physics of the interaction between energy and matter but also on the technology to acquire, process and display this huge quantity of biological data.
Not many years ago, medical imaging mainly referred to X-rays transmitted through the body and processed on film. In the last three decades however, imaging technology has grown rapidly, primarily as a result of the huge advances in digital computer technology, which continues to become less expensive, more compact and computationally faster.
Now, we can generate medical images of the human body in many different and more sophisticated ways, including the creation of moving 3D images. The intention of this presentation is to delve into the most common medical imaging techniques and explore how they are used to see inside the human body.
This event is organised by the IET Beds and Herts Branch.
Teresa Robinson, Bristol Royal Infirmary (IET Beds and Herts Event)
Venue: 7.00pm (with refreshments available from 6.30pm) Mitchell Hall, Cranfield University.
-
Real-time Adaptive Histogram Equalisation of Endoscopic Images for Medical Diagnosis - 22nd February 2008
Accurate and early endoscopic diagnosis demands a precise assessment of mucosal detail. The choice of techniques available to the clinician is often related to the availability of endoscopic equipment since many of the new technologies are device-specific. Our objective has been to devise a system that will generate real time, contrast-enhanced images to provide diagnostic information in a device and platform independent way.
Prof. Jonathan Brown, (Gloucestershire Royal Hospital / Cranfield University)
Venue: Red Lecture Theatre, Building 52, Cranfield University
-
Advancements in Machine Vision Software - 29th February 2008
This presentation will cover some of the most recent advancements which have been made with image processing development libraries, including colour analysis, 3D inspection and GPU (graphics processing unit) image processing. This seminar will especially focus on one of the industries leading image processing libraries, the Matrox Imaging MIL software development kit.
Matt Warren, Software Engineer (MVD Vision Solutions Ltd)
Venue: White Lecture Theatre, Building 52, Cranfield University
2007
-
Industrial Vision Systems : the theory and practice - 2nd February 2007
The theory behind industrial vision systems is very sound. The practice is sometimes very different. This seminar will attempt to reinforce some of the techniques used in vision systems but at the same time illustrate the pitfalls involved in dealing with real production line items.
Mike Phillips, CEO (Envisage Systems Ltd) & Martin Golding (Lambda Photometrics)
-
Developments in search and recognition algorithms - 9th February 2007
There are many approaches to pattern recognition - all have advantages and disadvantages. This seminar discusses several existing approaches together with a new technology that learns the key attributes of an image and then uses these to recognise future examples. The result is a technique that performs well with both fixed manufactured parts but also organic based tasks such as gender classification, texture recognition etc.
Jon Vickers, Technical Manager (Firstsight Vision)
-
Comparing Pattern Matching Techniques - 23rd February 2007
There are several methods available to locate a particular object within an image. This seminar will look at Normalised Greyscale Correlation (NGC) and Edge Based (GMF) pattern matching techniques, discussing the pros and cons of each method, in addition to presenting how to implement either algorithm.
Matt Warren, Software Engineer (MVD Vision Solutions Ltd)
-
High Speed PC based vision systems - 26th April 2007
How to create a leading edge vision application using commodity components - no programming required. This presentation will demonstrate how a very fast inspection system was created using standard technology - firewire, Windows XP and an 'open source' vision software. The focus will be on OCV - Optical Character Verification using template matching technology.
Paul Wilson (Scorpion Vision Ltd)
The Applied Mathematics & Computing Group are grateful to all the industrial speakers and the UK Industrial Vision Association (UKIVA) for their help in making the MSc industrial seminar series a success.