Improved Edge Detection by the Evaluation of Color Contrast Information.pdfVIP

Improved Edge Detection by the Evaluation of Color Contrast Information Maik Bollmann, Thorsten Hempel, Bärbel Mertsching University of Hamburg, Department of Computer Science, Section IMA Vogt-Kölln-Str. 30, 22527 Hamburg, Germany e-mail: bollmann@ or mertsching@informatik.uni-hamburg.de URL: rmatik.uni-hamburg.de Abstract The IMA Research Group has developed an active vision system, called NAVIS, which is mainly based on neural models. It can roughly be separated into five functional components: camera control, form detection, attractivity calculation, attention selection, and object recognition. The form module initially consisted of a gray value channel for the computation of luminance contrast. In this paper we describe the extension of the gray value channel by two color opponent channels which contribute to the extraction of edges. With these new channels it is possible to detect isoluminant edges. Furthermore, our model achieves a new estimation of the edges caused by luminance and color contrast. The attention and recognition capabilities of NAVIS can be improved by this method as experimental results show. 1 The Neural-Active-Vision-System NAVIS NAVIS is a biologically motivated active vision system designed to recognize complex objects within a non-uniform environment. Object recognition is done in a multistage fashion by first hypothesizing the presence and location of an object and afterwards identifying the object by its parts. A computer controlled camera mounted on a pan-tilt unit is used as the experimental platform for the evaluation of proposed concepts. An overview of the NAVIS architecture is presented in fig. 1. It shows the five functional components for camer