Dr. Shmuel Rippa received Ph.D. in mathematics from Tel-Aviv University in 1990. After working for three year as a research fellow in the Department of Applied Mathematics and Theoretical Physics (DAMTP) in the University of Cambridge, Shmuel joined Orbotech in 1993 where he filled various development positions. Shmuel Received the title of Orbotech Fellow In 2010 and he currently works on problems related to scientific computation, computational geometry and machine learning.
A Geometrical Descriptor Framework (GDF) for Filtering Anormalities in Polygonal Maps
Orbotech needs processing large polygonal maps. Sometimes the polygonal maps contains anomalities that are easy to observe by a human inspector but very hard to detect automatically. A natural approach for filtering out such anomalities is to register the polygonal map to a reference of a perfect panel and to filter out the differences. The problem is that high level of anomalous noise causes the registration process to fail. In this talk I will propose a procedure for detecting anomalities in the polygonal map without using a reference. The framework is composed of four steps: (1) Decomposition of the polygonal map into contours of equal arc-length, (2) Computing one or more descriptors for each contour, (3) Classification of the contour as either good or bad (noisy) and (4) semantic filtering of the pure noisy contours. I will describe the procedure, demonstrate its effectiveness on some examples and point to possible extensions.