Gudrun Socher

Bayesian reasoning on qualitative descriptions from images and speech

Image understanding denotes not only the ability to extract specific, non-numerical information from images, but it implies also reasoning about the extracted information. We propose a qualitative representation for image understanding results, which is suitable for reasoning with Bayesian networks. Our qualitative representation is enhanced with probabilistic information to represent uncertainties and errors in the understanding of noisy sensory data. The probabilistic information is supplied to a Bayesian network in order to find the most plausible interpretation. We apply this approach for the integration of image and speech understanding in a scenario where we want to find objects in a visually observed scene which are verbally described by a human. Results demonstrate the performance of our approach.

Motion boundary detection in image sequences by local stochastic tests

While estimating both components of optical flow based on the postulated validity of the Optical Flow Constraint Equation (OFCE), it has been tacitly assumed so far that the partial derivatives of the gray value distribution — which are required for this approach at the pixel positions involved — are independent from each other. [Nagel 94] has shown in a theoretical investigation how dropping this assumption affects the estimation procedure. The advantage of such a more rigorous approach consists in the possibility to replace heuristic tests for the local detection of discontinuities in optical flow fields by well known stochastic tests. First results from various experiments with this new approach are presented and discussed.

Multilevel Integration of Vision and Speech Understanding Using Bayesian Networks

The interaction of image and speech processing is a crucial property of multimedia systems. Classical systems using inferences on pure qualitative high level descriptions miss a lot of information when concerned with erroneous, vague, or incomplete data. We propose a new architecture that integrates various levels of processing by using multiple representations of the visually observed scene. They are vertically connected by Bayesian networks in order to find the most plausible interpretation of the scene. The interpretation of a spoken utterance naming an object in the visually observed scene is modeled as another partial representation of the scene. Using this concept, the key problem is the identification of the verbally specified object instances in the visually observed scene. Therefore, a Bayesian network is generated dynamically from the spoken utterance and the visual scene representation. In this network spatial knowledge as well as knowledge extracted from psycholinguistic experiments is coded. First results show the robustness of our approach.

3-D reconstruction and camera calibration from images with know objects

We present, a method for camera calibration and metric reconstruction of the three-dimensional structure of scenes with several, possibly small and nearly planar objects from one or more images. We formulate the projection of object models explicitly according to the pin-hole camera model in order to be able to estimate the pose parameters for all objects as well as relative poses and the focal lengths of the cameras. This is accomplished by minimising a multivariate non-linear cost function. The only information needed is simple geometric object models, the correspondence between model and image features, and the correspondence of objects in the images if more than one view of the scene is used. Additionally, we present a new method for the projection of circles using projective invariants. Results using both simulated and real images are presented.

Talking about 3D scenes: integration of image and speech understanding in a hybrid distributed system

We present a hybrid system that integrates speech and image understanding. Given spoken references, it is able to identify objects of a 3D scene perceived via a stereo camera. Central to our approach is the extraction of qualitative object features and spatial scene properties from acoustic and visual data. The interaction of the understanding processes is performed using a procedural semantic network that interfaces with signal recognition and reconstruction modules, thus integrating semantic, neural and Bayesian networks and Hidden Markov Models.

Koordinatensysteme und Transformationen

Im vorhergehenden Kapitel wurde gezeigt, wie man geometrische Objekte in der 3DComputergrafik modelliert. In diesem Kapitel wird dargestellt, wie man diese Objekte in einer Szene positioniert, wie man die Position und die Blickwinkel einer Kamera festlegt, die die Szene quasi fotografiert, und wie man schlieslich die Ausmase des fertigen Bildes spezifiziert, das in einem Fenster des Bildschirms dargestellt werden soll. All diese Aktionen werden durch entsprechende Koordinatentransformationen erreicht.

Ellipsenbasierte 3-D Rekonstruktion

In dieser Arbeit wird ein Verfahren zur Schatzung der dreidimensionalen Lage von Kreisflachen vorgestellt. Die Ausnutzung projektiver Invarianten ermoghcht eine eindeutige Losung fur eine auf einem Stereobild abgebildete Kreisflache bei bekannten Kameraparametem und bekanntem Radius. Aus zwei projizierten Kreisflachen mit bekanntem Radius kann neben der eindeutigen Lage auch die relative Lage der Kameras geschatzt werden, wenn die Kreisflachen komplanar sind oder ihre relative raumliche Lage bekannt ist.

3D visualization of legislative relationships

Government relationships can be complex and difficult to understand. The relationships between members of a legislature, bills, votes and lobbyists who promote various causes are important to understand in representative democracies, but difficult to retrieve using current methods. In this paper, we propose a 3D visualization system to explore such legislative relationships for users. We use real data from California state legislature obtained from the Digital Democracy project. We also conduct a 20 person user study to gauge the differences between traditional ways of looking up information versus our graph based methods. While not being as comprehensive, most users found our interactive visualizations more intuitive than regular web-based information retrieval.

POSTER: Security Design Patterns With Good Usability

This poster presents work-in-progress in the field of usable security. The usability of security mechanisms is crucial to avoid unintended misuse of security mechanisms which lowers the security level of a system. It is the goal of the work presented in this poster to identify security design patterns with good usability. Requirements for security design patterns with good usability stem from existing usable security design guidelines. A collection of security usability failures is presented as well as examples of how misuse anti-patterns can be derived from these failures. Misuse cases will be used in future work to identify security design patterns with good usability.

Zusammenhang zwischen Computergrafik und Bildverarbeitung

Warum fasst man die Gebiete Computergrafik und Bildverarbeitung in einem Buch zusammen? Weil sie die zwei Seiten einer Medaille sind: wahrend man in der Computergrafik aus einer abstrakten Objektbeschreibung ein Bild generiert, versucht man in der Bildverarbeitung nach der Extraktion von charakteristischen Merkmalen die im Bild enthaltenen Objekte zu erkennen und so zu einer abstrakten Objektbeschreibung zu kommen (Bild 2.1). Oder anders ausgedruckt: Computergrafik ist die Synthese von Bildern und Bildverarbeitung ist die Analyse von Bildern. In diesem Sinne ist die Computergrafik die inverse Operation zur Bildverarbeitung.