题目:Gestalt Algebra
报告人:Dr. Eckart Michaelsen, Fraunhofer Institute of Optronics, System Technologies and Image Exploitation, Germany
Gestalt Algebra captures the laws of perceptual grouping in human vision as they have been investigated for much more than hundred years now by psychologists. It is meant to be used in machine vision as complement of machine learning and automatic utilization of knowledge. Three operations are defined for this algebra – one for mirror symmetry, one for good continuation in rows, and one for rotational symmetries. It allows a deep decomposition of objects found e.g. on remotely sensed imagery into a hierarchy of parts which are given in meaningful mutual arrangements. In contrast to picture grammars the algebra may not contain threshold parameters, it has no constraints. Instead the mutual relations are captured by continuous functions assigning assessments to the terms of the algebra. Image analysis with this structure starts by extracting a set of primitive Gestalten from the image. Then the search for composite Gestalten assessed as most meaningful begins. For the time being mostly bottom-up search is discussed.
Eckart Michaelsen was born in 1958 in Stade, Germany. He graduated from the University of Innsbruck (Austria) in 1987 with diploma on Mathematics. He started working for Forschungsinstitut für Mustererkennung, Forschungsgesellschaft für angewandte Naturwissenschaften (FIM-FGAN) in Ettlingen (Germany) the same year, and stays with this affiliation since now. Today this lab is integrated in the IOSB (Institut für Optronik, Systemtechnik und Bildauswertung) of the Fraunhofer Gesellschaft. In 1998 Eckart Michaelsen received Dr.-Ing. from the University of Erlangen (chair for pattern recognition H. Niemann), working on syntactic methods of pattern recognition. He holds the chair of IAPR Technical Committee TC7 (pattern recognition in remote sensing and mapping) since 2014. He is associate editor of the Pattern Recognition Letters and the Pattern Recognition and Image Analysis journal. Occasionally he serves as reviewer and opponent for PhD defenses such as lately for the CMP-FELK laboratory of the Czech Technical University in Prague. Michaelsen published over 90 scientific papers mostly on knowledge-based analysis of remotely sensed images, Gestalt perception in machine vision and automatic visual UAV navigation in peer reviewed journals and conferences. He is member of program committees for conferences such as ICPR, ISPRS, IGARS, PRIA, and CAIP.

Presented by Dr. Eckart Michaelsen