Showing papers by "Gerhard Sagerer published in 1995"

A communication framework for heterogeneous distributed pattern analysis

Abstract: Unlike in traditional approaches to parallel or distributed processing where normally well structured problems are implemented completely in some programming environment we are faced with the problem of integrating existing heterogeneous software systems. Furthermore, pattern analysis stresses special aspects of communication capabilities. Therefore, we propose a new communication framework dedicated to heterogeneous pattern analysis systems that handles typed structured data, enables completely symmetric interaction, and provides various call semantics. A first prototype evaluating some of the concepts in practical situations is presented. >

Segmentation of molecular surfaces based on their convex hull

Abstract: Docking of two or more proteins to protein complexes is based on the complementarity of shape and chemical attributes of the involved molecular surfaces. A geometry-based segmentation technique is proposed, generating characteristically shaped regions of molecular surfaces, used to predict possible docking sites of proteins. Thus, an unrestricted search in 6D configuration space is avoided. The segmentation algorithm is based on the convex hull of the molecular surface. Its implementation has been shown to produce stable segmentations of the 3D surfaces, well approximating the true contact sides of the involved proteins. Making as few as possible assumptions on the underlying surfaces this technique might be useful to other applications as well.

Protein docking combining symbolic descriptions of molecular surfaces and grid-based scoring functions

Abstract: With the growing number of known 3D protein structures, computing systems, that can predict where two protein molecules interact with each other is becoming of increasing interest. A system is presented, integrating preprocessing like the computation of molecular surfaces, segmentation, and searching for complementarity in the general framework of a pattern analyzing semantic network (ERNEST). The score of coarse symbolic computations is used by the problem independent control strategy of ERNEST to guide a more detailed analysis considering steric clash and judgements based on grid-based surface representations. Successful examples of the docking system are discussed that compare well with other approaches.

Generation of Language Models Using the Results of Image Analysis

Abstract: USING THE RESULTS OF IMAGE ANALYSIS Uta Naeve Gudrun Socher Gernot A. Fink Franz Kummert Gerhard Sagerer Universit at Bielefeld, Technische Fakult at, AG Angewandte Informatik, Postfach 100131, 33501 Bielefeld, Germany, e-mail: uta@techfak.uni-bielefeld.de ABSTRACT We present a new approach towards using contextual information to enhance speech recognition and understanding. Dynamically inferred knowledge about the context is used in addition to the static linguistic and domain speci c knowledge. Based on the results of image analysis of a given scene language models for constituents of possible utterances concerning that scene are generated.

Detection of unknown words and its evaluation.

Abstract: Especially in recognition of spontaneous speech it is necessary to be able to cope with the occurance of unknown words. We present an approach to unknown word detection integrated with speech recognition. Though several schemes exist for the evaluation of detection algorithms all suffer from some deficiencies. Therefore, we also present a new evaluation measure called detection accuracy which is similar to the widely accepted word accuracy. We applied this measure to the evaluation of our approach on a large spontaneous speech recognition task from the German Verbmobil project.

Statistical classification and segmentation of biomolecular surfaces

Abstract: A preprocessing algorithm is presented, integrating the computation of biomolecular surfaces, the calculation of geometrical and chemical features on each surface point, and a complex segmentation into characteristic surface regions. Calculated regions cover the whole computed protein surface and are a feature of shape, roughness and hydrophobicity on the appropriate surface area. Such regions can be paired in complementary parts, used to predict possible docking sites of proteins. The segmentation algorithm of docking sites proceeds in two steps: first, surface points are classified with a vector quantizer. Second, connected areas of surface points are computed by a complex region growing technique. The detected regions were compared to the real docking sites calculated from complexes with known 3D structures with good results of approximation.

Representing Procedural Knowledge for Semantic Networks using Neural Nets

Abstract: Interpretation of complex scenes involves analysing multiple objects being composed of several parts. Since diierent objects often have parts in common it is useful to share resources representing these identical substructures. This is quite a diicult task for artiicial neural networks (ANNs) 4], but can be handled with semantic networks. They are a well established tool for representing and organizing scene models 1, 7]. However, signal interpretation needs to be robust against distortions and ad-aptive in diierent environments. These properties are characteristic advantages of ANNs 12] which suggests to combine them with semantic nets. In this paper we propose a combination of ANNs and semantic networks attempting to combine the beneets of both approaches. First results on visual recognition of composed objects are promising.

Automatische Detektion von Wurzelsystemen in Minirhizotron-Bildern

Abstract: An der Wurzelforschung besteht aufgrund der vielfaltigen Wechselwirkungen zwischen Wurzelsystem und Boden ein groses Interesse. Fur aussagekraftige Untersuchungen ist die Analyse von grosen Bildmengen erforderlich. Die manuelle Auswertung ist zeitaufwendig und die Qualitat der Ergebnisse hangt sehr stark von der Erfahrung der auswertenden Person ab. Dieses fuhrt zu einer zunehmenden Automatisierung der Bildauswertung. Jedoch enthalten die Bilder komplexe und zufallig verzweigte Wurzelsysteme (s. Abb. 2). Die Detektion der Wurzelregionen ist besonders schwierig, da der Hintergrund sehr stark variiert und der Kontrast zwischen Boden und Wurzel oft sehr gering ist. Alle existierenden Verfahren zur automatischen Auswertung dieser Bilder sind regionenbasierte Ansatze, die davon ausgehen, das die Grauwerte des Wurzelsystems sich gut von den Grauwerten des Hintergrundes trennen lassen. Diese Annahme ist aber im allgemeinen nicht erfullt. Aus diesem Grund verfolgen wir in dieser Arbeit einen konturorientierten Ansatz. Nach der Detektion von signifikanten Wurzelbereichen, erweitern wir diese mittels heuristischer Suche zur optimalen Kontur. Die Basis des Verfahrens bildet der A—Algorithmus. Bei der Suche wird die parallele Struktur der Wurzel ausgenutzt, d.h. die Informationen der rechten und der linken Kontur fliesen gemeinsam in die Bewertung ein.

Robuste inkrementelle Verarbeitung gesprochener Sprache

Abstract: Der von uns entwickelte Ansatz zur inkrementel- len Verarbeitung gesprochener Sprache basiert auf einem flexiblen daten- imd erwartimgsgesteuerten Analyseverfahren und erlaubt die Integration von Spracherkennung und Sprachverstehen. Diese Vorgehensweise stellt spezielle Anforderungen an die Robustheit der Analysestrategie. Nicht mit der Wissensbasis konforme Teile einer Auserung werden mit Hilfe spezieller akustischer Modelle detektiert. Auserdem wird dynamisch aufgrund der Suchbaumentwicklung eine inefiiziente Breitensuche verhindert. Diese Strategie wird anhand einer spontansprachlichen Stichprobe evsluiert.