by Joost Batenburg (CWI) and Tamas Sziranyi (MTA SZTAKI)
For the young generation it is hard to imagine that just two decades ago, taking a picture took days, or even weeks, requiring us to wait until the film was full, after which it was taken to a photo shop for further development. These days, digital cameras are all around us, which has revolutionised the way we deal with images. The development of digital sensors has followed a similar path in other disciplines within science and engineering, resulting in the development of a broad range of detectors and sensors that can collect various types of high-dimensional data reflecting various properties of the world around us. This has fuelled the development of a new field of mathematics and computation, which deals with interpreting such sensor data, applying algorithms to it, and generating new data.
Image understanding has been in relation to several disciplines, depending on the purpose of the analysis, for example, saliency, scale-space theory or rules of the real world.
When it comes to interpreting human sensation and augmented visibility, saliency analysis and psychophysiology play an important role. Computer vision can be described by a few axioms, like invariant transformations. This theory leads to a coherent discipline of scale-space theory, yielding beautiful and very effective tools such as anisotropic diffusion or SIFT features. However, computer vision is a projection of the real world. To describe this world well, we must consider its laws: laws of physics, laws of biology, laws of optics and laws of chemistry. This means that not only pixels and pixel formations are considered in one or more image planes, but we must search for other relationships among objects representing real artefacts. This modelling of the real world sometimes needs huge computing power, while the sophisticated programming techniques of such cases necessitate careful computational analysis. As better sensors and more powerful computers are being developed, simple image processing is evolving into the field of computational imaging.
The field of computational imaging colligates a broad range of imaging techniques where computation plays an integral role in the image formation process. By its very nature, the field of computational imaging is highly interdisciplinary. Firstly, the process by which the input data for a computational imaging approach are formed often involves physics, which must be modelled to properly interpret the data. Many computational imaging techniques also involve the design and implementation of sophisticated sensor systems that involve a combination of physical and computational aspects. Secondly, the way the actual images are formed usually involves mathematics to derive the transformations that are applied to transform the input data into the desired image form, and analyse its properties. In particular, the field of inverse problems plays a crucial role here, as we often want to create an image of some object that can only be observed through indirect measurements. Finally, computer science is crucial for the creation of efficient, scalable algorithms and software implementations that can deal with large-scale image data. It involves aspects from high-performance computing, database technology, and software engineering for dealing with the complex requirements of the broad range of users of such techniques.
Applications of computational imaging can be found in many different fields: it is a crucial tool for scientific research of biomedical systems and advanced materials, making it possible to perform quantitative measurements based on images collected using microscopes and other imaging devices. On a very different scale, images of the earth, both its surface and interior, are computed from sensor measurements in geoscience and remote sensing applications. Often this data is further processed by sophisticated image analysis techniques that perform feature recognition and classification.
Despite the broadness of the field and large variety of mathematical and computational techniques, several common trends can be identified across the application domains involved in computational imaging. Driven by improvements in mathematical modelling and computational capabilities, multi-modal and multi-channel imaging is rapidly gaining importance. By combining sensor information from different sources, or using multiple energy channels, complex image models can be formed that can be digitally visualised and analysed to extract meaningful quantitative information. Also, the ability to collect imaging data at an unprecedented rate and scale is bringing up new challenges to deal with the massive scale of big data problems in imaging.
Each of the articles in this issue of ERCIM News touches upon one, or several, of the topics mentioned above. Collectively they provide an insight in the many complex facets of computational imaging that are currently being investigated.
The papers received for this special theme can be grouped into some topics that characterise the activities of the ERCIM research teams:
Biological and medical imaging: making the inside visible
Convolutional Network Layers Map the Function of the Human Visual Cortex
by Michael Eickenberg, Gaël Varoquaux, Bertrand Thirion (Inria) and Alexandre Gramfort (Telecom Paris)
Data Fusion at the Nanoscale: Imaging at Resolutions Better than Wavelength/100
by Bernd Rieger and Sjoerd Stallinga (Delft University of Technology)
Computational Processing of Histological Images
by Erwan Zerhouni, Bogdan Prisacari, Maria Gabrani (IBM Zurich) and Qing Zhong and Peter Wild (Institute of Surgical Pathology, University Hospital Zurich)
Segmenting Cells in the Presence of a Diffuse and Heterogeneous Membrane Marker
by Christophe De Vleeschouwer (UCL) and Isabelle Migeotte (ULB)
Application of Digital Holographic Microscopy for Automatic Monitoring of Freely Floating Microorganisms
by László R. Orzó, Zoltán Á. Varecza, Márton Zs. Kiss and Ákos Zarándy (MTA SZTAKI)
Modelling Neurodegenerative Diseases from Multimodal Medical Images
by Olivier Colliot (CNRS), Fabrizio De Vico Fallani and Stanley Durrleman (Inria)
Microwave Imaging for Brain Stroke Detection and Monitoring using High Performance Computing
by Pierre-Henri Tournier (Inria)
Geology and remote sensing: observation of Earth from above and looking below the surface
Natural Disaster Monitoring: Multi-Source Image Analysis with Hierarchical Markov Models
by Josiane Zerubia (Inria), Gabriele Moser and Sebastiano B. Serpico (University of Genoa)
Processing Satellite Imagery to Detect and Identify Non-collaborative Vessels
by Marco Reggiannini and Marco Righi (ISTI-CNR)
Geometric Imaging for Subsurface Salt Bodies
by Tristan van Leeuwen (Utrecht University), Ajinkya Kadu (Utrecht University) and Wim A. Mulder (Shell Global Solutions International B.V. / Delft University of Technology)
Multi-view: 3D geometry fraps the camera views
Bringing Modern Mathematical Modelling to Orientation Imaging for Material Science
by Nicola Viganò (CWI)
Towards Computational Photomechanics
by Frédéric Sur (Université de Lorraine, Inria), Benoît Blaysat and Michel Grédiac (Université Clermont-Auvergne)
Novel Concepts for Wave Imaging in Complex Media
by Lorenzo Audibert (EDF) and Houssem Haddar (Inria)
Multi-spectral: Colour channels describing the material insight
Low-dose X-ray Tomography Imaging Based on Sparse Signal Processing
by Samuli Siltanen (University of Helsinki)
3D Flashback: An Informative Application for Dance
by Rafael Kuffner dos Anjos, Carla Fernandes (FCSH/UNL) and João Madeiras Pereira (INESC-ID)
Identifying Persons of Interest in CCTV Camera Networks
by Furqan M. Khan and Francois Bremond (Inria)
Computational Fusion of Multi-View and Multi-Illumination Imaging
bySvorad Štolc, Reinhold Huber-Mörk and Dorothea Heiss
Multispectral Imaging for the Analysis of Cultural Heritage Objects and Image Registration for Results Optimisation
by Kostas Hatzigiannakis, Athanasios Zacharopoulos and Xenophon Zabulis (ICS-FORTH)
Computational Snapshot Spectral Imaging
by Grigorios Tsagkatakis and Panagiotis Tsakalide
Lip Segmentation on Hyper-Spectral Images
by Alessandro Danielis, Daniela Giorgi and Sara Colantonio (ISTI-CNR)
Other
CV-HAZOP: Introducing Test Data Validation for Computational Imaging
by Oliver Zendel, Markus Murschitz and Martin Humenberger (AIT Austrian Institute of Technology)
Please contact:
by Joost Batenburg
CWI, The Netherlands
Tamas Sziranyi
MTA SZTAKI, Hungary
