An evaluation of classifiers for reading resistor colors

A lot of effort has been devoted to reading resistor colors using image processing and pattern recognition techniques. It is not so clear which classifier or machine learning is effective for classifying colors in reading a resistance of a resistor. This paper presents an evaluation of classifiers for reading resistor's colors on an RGB color space under various illumination situations. Eight classifiers to be examined are k-nearest neighbor (k-NN) (k=1, 3, and 5), decision tree (DT), support vector machine (SVM), Gaussian naive Bayes (NB), artificial neural network (ANN), and random forest (RF). The classification performance of 8 classifiers is evaluated by the average error rate, respectively. From the experimental results, depending on the training sample size and illumination situations, the classifier to be used for reading resistor colors should be considered. Considering practical color pattern recognition problems with poor illumination conditions, the 1-NN classifier should be the more practical and usable classifier. This study will provide one of the ways for AI and robotics applications to accurately classify colors.