Joint Transform Correlator Based on Color Human Vision Model for Polychromatic Pattern Recognition

摘要

Electro-optical system that combines the advantages of optical and electronic system has been widely applied in signal processing. Because of advantages of real-time, parallel operation capability, and programmable flexibility, it is especially utilized in the pattern recognition, e.g., military radar system, secure entry system, biomedical detection, and many other applications. Moreover, color is one of the most significant and powerful visual information. Improving the JTC for polychromatic pattern recognition is a novel topic. Deutsch proposed a multi-channel single-output JTC. It is based on RGB color model and separates the color pattern into red, green, and blue components respectively and then displays on the spatial light modulator (SLM). Thereafter, RGB become a popular model to perform pattern recognition. However, it is known that red, green, and blue components are correlated with one another. Furthermore, there are some cases in which the use of the RGB chromatic system is not appropriate for obtaining effective polychromatic object recognition, e.g., RGB decomposition is not appropriated for discriminating between similar color objects. Therefore, we propose a novel nonzero order JTC based on ATD color model to achieve polychromatic pattern recognition. ATD color model is constituted by achromatic, tritanopic, and deuteranopic systems. People consider that ATD color model is the most to be close to human eyes. In this paper, we apply ATD color model to multi-channel nonzero order joint transform correlator and perform pattern recognition. To estimate whether ATD color model is suitable for pattern recognition, we compare with RGB, which is common used to perform pattern recognition. The terms of estimation contain recognition ability of rotational distortion, brightness performance, multi-level quantized reference functions, noise tolerance ability, and relationship between channels and channels. We utilize peak to correlation energy, peak to sidelobe ratio, correlation peak intensity, and mutual correlation coefficient as our performance evaluation parameters. Finally, the results show that ATD color model is suitable for pattern recognition.