Reaction-Diffusion Algorithm for Vision Systems
Atsushi Nomura, Makoto Ichikawa, H. Rismon, Hidetoshi Miike · 2007
This chapter presented the reaction-diffusion algorithm for vision systems. After a brief explanation of the reaction-diffusion system, we presented a class of algorithms for edge detection, grouping and stereo disparity detection by utilizing the FitzHugh-Nagumo type reaction-diffusion equations; all of the algorithms are necessary for the realization of vision systems. Previous algorithms, in particular, those proposed by Marr and his collaborators, utilize the Gaussian filter; the output of the filter is equivalent to the solution of the diffusion equation. In contrast to this, the reaction-diffusion algorithm has non-linear reaction terms coupled with diffusion equations. The non-linearity of the algorithm and the Turing-like condition can help to achieve good performance in edge detection, grouping and stereo disparity detection. Recently, the authors found a key mechanism in the stochastic resonance for performance improvement (Ebihara et al., 2003b). Thus, we conclude this chapter by noticing that further performance improvement will be possible with the use of the stochastic resonance in the reaction-diffusion algorithm.