Obiettivo
Computer vision tasks are normally attacked in one of two different angles: bottom-up or top-down. The first approach has the problem of sorting out details to pass to higher levels without filtering out necessary information for a given task. The latter comprises a major computational burden. We propose to solve computer vision tasks at all levels simultaneously using the deep scale-space structure of images. It is the thesis of this consortium that computer vision tasks may be elegantly, robustly and efficiently addressed via the multi-scale singularity structure of images. We will investigate the underlying mathematics and the necessary tree and graph algorithms so as to draw conclusion on applicability in areas such as image databases, medical imaging, and image coding. Furthermore, the consortium will apply this methodology to solve specific tasks in medical imaging.
OBJECTIVES
We will develop theory & practice in singularity theory, scale-space theory & algorithmic to create efficient algorithms for solving computer vision (CV) tasks. The overall objective is to summarise images in a generic tree/graph data structure, efficiently sub serving CV tasks. We will investigate alternative representations & their expressive power. New representations of images suitable for storage & further computation may be created. We will investigate how these data structures sub serve CV tasks e.g. image coding, shape representations, structural search, matching, indexing. They comprise the core of CV tasks e.g. image communication, database search, registration of images, computer aided diagnosis, object recognition. We will represent anatomical objects in 3D CT&MR images of the craniofacial region, compare similar objects & index objects in 3D image databases.
DESCRIPTION OF WORK
We will attack the problem of finding suitable multi-scale image summaries from three different angles: singularity theory, algorithmic, and scale-space theory. Singularity theory will be used for defining relevant structures in multi-scale images. We will investigate both the representation of 3D shape and the singularities in scale-space. We will solve so far open problems in the structure of 3D skeletons and multi-scale singularity structure of features of images. In algorithmic, we will investigate the computational complexity of solving computer vision tasks using tree and graph algorithms.
Furthermore, we will develop new algorithms of lower complexity in the specialized case of certain computer vision tasks. Finally, we aim at transferring knowledge from the algorithmic community to the computer vision community about solutions, and vice versa about relevant problems. In scale-space theory, we will investigate the expressive power of multi-scale structures of images. This is in terms of top points, edges, blobs, ridges, and especially their generic multi-scale topological structure. We will apply these results to generic computer vision tasks, and we will apply them to specific problems in storage of medical images, computer aided diagnosis, and retrieval of medical images from large databases. Finally, the algorithms to be developed will be evaluated on two databases to be collected, one of 3D images of the craniofacial area and one of teeth. The consortium will as an infrastructure for the project, and for the scientific community, develops tools for visualization and fast prototyping.
Campo scientifico
Argomento(i)
Invito a presentare proposte
Data not availableMeccanismo di finanziamento
CSC - Cost-sharing contractsCoordinatore
2300 COPENHAGEN S
Danimarca