Multichannel pattern recognition of color images

In this paper we present the research line that we develop in pattern recognition of multichannel images, centered in the application of color images. Pattern recognition is performed through a multichannel correlation process. The correlation is applied to each channel, red, green and blue of the color image. The final recognition result is obtained by a combination of the information of the three monochromatic correlations. Two different approaches are proposed in order to improve the discrimination capability of the multichannel process. First, element-wise transformations over the multichannel images are used in order to enhance differences between channels. Then, the information in each channel is independent and the autocorrelation is enhanced with respect to the cross- correlations. The second approach involves the optimization of the matched phase-only filters used in every channel. This optimization is performed by means of a region of support. They are two complementary techniques that increase the discrimination capability and eliminate false alarms. The result is a better performance of the multichannel correlator for pattern recognition.