Multidimensional image data analysis
Analysis of 3D, 4D or higher dimensionality image data, such as single and dual energy CT, MRI, PET. These volumetric data are the quickest rising data sets and not easy for human perception as solid 3D or 4D texture is nothing that humans can see dorectly but soemthing that requires visualization.
Information retrieval evaluation
Evaluation and benchmarking of information retrieval methods, mainly automatic but also user centered strategies. ImageCLEF has created a platform for such benchmarking in several domains.
Multimodal information retrieval and information fusion
Multimodal retrieval includes the cobination of several media for retrieval, for example text and images or images and structured data with the goal to combine all available information for optimal retrieval and combining the results.
Medical image analysis and retrieval
Medical image analysis and retrieval has several differences from general retrieval. Most often only very small parts and structures are really relevant for the retrieval, and thus loal features are required. Most often a combination with other data suchas age and gender are necessary for good retrieval. Greay levels and textures are often more important than colors for retrieval.
Test collection creation including signals and images
Test collections are the basis for evaluation in many research domains. Acquiring data in a very standardized way and keeping the goal in mind at the data acquisition time is important to ensure the validity of the results. Our test collections are mainly medical and include acquisitions of electro-myography (EMG) data, data from body movement sensors, data from force strength sensors, image data sets and (for the medical datasets) also clinical parameters that can be important for interpreting signals and images.
User testing and task analysis
Working with real users is often not easy and user tests can lead to many problems when not well prepared.
Different users are usually assumed to be similar, but they can be very different (as when acquiring electro-myographyc signals from amputated and non amputated subjects). Moreover, they always behave in slightly different ways during the tests, and they are often acquired in different experimental conditions. Instead, different tasks are usually thought in order be very different one from another, and to stress different parts of the experimental setting.
Therefore, user centered evaluation depends on causes that are often far from the ones that determine system-oriented evaluation.
However, both of the causes usually lead to important data features that must be analyzed deepely and separately.
Infrastructures for computation
Quickly increasing data sets have led to many computing challenges in domains such as bioinformatics, simulation and image processing. Grid networks allow resources sharing and also Clouds and virtualization can help to facilitate managing computing resources. Where the bottleneck is in terms of processor power, bandwidth or storage needs to be analyed to work on the best possible infrastructure for research.
