INSTITUTE OF PHYSICS PUBLISHING NETWORK: COMPUTATION IN NEURAL SYSTEMS Network: Comput. Neural Syst. 13 (2002) 1–23 PII: S0954-898X(02)53197-3 Processing NET/net153197-xsl/PAP Printed 19/9/2002 Focal Image (Ed: STUART ) CRC data File name NE .TEX First page Date req. Last page Issue no. Total pages Multi-modal estimation of collinearity and parallelism in natural image sequences* Norbert Kr ¨ uger and Florentin W¨ org¨ otter Institute for Computational Intelligence and Technology (INCITE), University of Stirling, Scotland FK9 4LA, UK E-mail: norbert@cn.stir.ac.uk Received 9 January 2001, in final form 22 August 2002 Published Online at stacks.iop.org/Network/13/1 Abstract In this paper we address the statistics of second-order relations of feature vectors derived from image sequences. We compute the individual vector components corresponding to the visual modalities orientation, contrast transition, optic flow, and colour by conventional low-level early vision algorithms. As a main result, we observe that collinear (or parallel) line pairs are, with very great likelihood, also associated with other identical features, for example sharing the same flow pattern, or colour or even sharing multiple feature combinations. It is known that low level processes, such as edge detection, optic flow estimation and stereo are ambiguous. Our results provide support for the assumption that the ambiguity of low level processes can be substantially reduced by integrating information across visual modalities. Furthermore, the attempt to model the application of gestalt laws in computer vision systems based on statistical measurements, as suggested recently by some researchers (Elder H and Goldberg R M 1998 Perception Suppl. 27; Geisler W S, Perry J S, Super B J and Gallogly D P 2002 Vis. Res. 41 711–24; Kr¨ uger N 1998 Neural Process. Lett. 8 117–29; Sigman M, Cecchi G A, Gilbert C D and Magnasco M O 2001 Proc. Natl Acad. Sci. USA 98 1935–49), gets further support and the results in this paper suggest formulation of gestalt principles in artificial vision systems in a multi-modal way. (Some figures in this article are in colour only in the electronic version) 1. Introduction Substantial research has been focused on the usage of gestalt laws in computer vision systems See endnote 1 (excellent overviews are given in [9, 62]). The most often applied gestalt principle in artificial visual systems and also the most dominant gestalt principle in the 2D projection of natural scenes is collinearity [14, 21, 45, 66]. Collinearity can be exploited to achieve more robust See endnote 2 * This work has, to a large part, been performed during Norbert Kr¨ uger’s stay in the Cognitive System Group at the University of Kiel, Germany. 0954-898X/02/000001+23$30.00 © 2002 IOP Publishing Ltd Printed in the UK 1