Multi-channel chromatic transformations for nonlinear color pattern recognition A. Fares a , P. Garc  ıa-Mart  ınez b, * , C. Ferreira b , M. Hamdi a , A. Bouzid a a D epartement de Physique, Facult e des Sciences, Laboratoire de Photonique, Universit e Moulay Ismail, B.P. 4010 Mekn  es, Morocco b Dept.  Optica. Universitat de Val  encia., c/Dr. Moliner, 50. 46100 Burjassot, Valencia, Spain Received 31 October 2001; accepted 10 January 2002 Abstract We present a new approach for color pattern recognition based on multi-channel nonlinear correlations. High discrimination capability is obtained in comparison with common linear multi-channel detection methods. We apply the nonlinear morphological correlation to different color channel decompositions as RGB and ATD channels. Moreover, in order to improve the discrimination we have introduced a new color transformation. When a high se- lectivity is required, the combination of the nonlinear correlation and the new color decomposition yields to detect the object using just a single channel. Simulation results are provided. Ó 2002 Elsevier Science B.V. All rights reserved. Keywords: Color image; Morphological correlation; Pattern recognition; Color transformations 1. Introduction Color is one of the most powerful and impor- tant information to be considered for image pro- cessing,inparticularforpatternrecognition.From the optical pattern recognition point of view, the linear correlation is one of the most useful tool [1]. In fact, optical correlation can be extended to polychromatic images. The first color detection process was introduced by Yu and Chao [2]. In that work, they used three coherent light sources (red, green and blue, RGB) and a diffraction grating. An extensive method to study color pat- tern recognition is the multi-channel correlator [3– 5]. With this approach the input scene and the color target are decomposed into three color channels and then processed separately. Finally, the correlation outputs are combined to give a fi- nal detection using arithmetic or logical point-wise operations. Much work has been devoted for the multi-channel processing in order to improve the process [6–9]. Nevertheless, color pattern recognition has not been studied only with multi-channel methods. There are methods based on preprocessing the input scene and the color target in order to com- bine the color information in a unique image. Those are the single-channel approaches and some work has been done based on its application to two-dimensional [10,11] and tri-dimensional im- 15 March 2002 Optics Communications 203 (2002) 255–261 www.elsevier.com/locate/optcom * Corresponding author. Tel.: +34-96-386-47-17; fax: +34-96- 386-47-15. 0030-4018/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved. PII:S0030-4018(02)01184-7