Color-blindness plate recognition using a neuro-fuzzy approach

The recognition of patterns in a colored environment is an interesting but complex problem. As a tool for testing the color-sensory ability of human eyes in a colored environment, color-blindness plates can be applied as test samples for pattern-recognition models for color images. In this research, a modified connectionist fuzzy classifier for recognition (MCFC-R) is developed for automatically recognizing patterns in color-blindness plates. The model consists of two subnetworks and a foreground image-reconstruction process between the two subnetworks. Both the subnetworks in the MCFC-R are derived from a connectionist fuzzy classifier (CFC) model, which realizes a fuzzy classification procedure in a massively parallel architecture [Kuo, Y.-H., Kao, C.I., Chen, J.-J., 1993. A fuzzy neural network model and its hardware implementation. IEEE Trans. Fuzzy Syst. 1(3), 171-183; Kuo, Y.-H., Hsu, J.-P., Kao, C.-I., 1996. MCFC: a fuzzy neural network model for speech recognition. J. Intell. Fuzzy Syst. 4, 257-268.]. The first subnetwork is responsible for the task of image segmentation based on color features. The second subnetwork is applied to the job of object recognition (numeric character recognition in this research). The reconstruction process between the two subnetworks includes image fuzzification, scanning window selection, feature extraction and foreground image decision procedures to extract the foreground image. According to the experimental results, the MCFC-R model displayed good performance in the recognition of color-blindness plates, even in a shift-variant, scale-variant or data-lost situation.