Chess pieces classification in images using convolutional neural networks

Classifiers of chess pieces in images are important part of many computer vision systems related to chess. Exist various implementations of chess pieces classifiers, one of which is convolutional neural networks. This paper examines validity of convolutional neural networks as chess pieces classifiers. This paper describes experiments, which compare different CNN architectures, training processes, compare achieved results with already existing classifiers and evaluate created models ability to classify chess pieces in various angles. This paper proves that CNN is a valid solution for chess pieces classification in images problem. Provided CNN architecture displays better results than analyzed classifiers used in other systems. This paper suggests how to minimize effect of overfitting when using small training dataset and a few potential methods to improve CNN ability to classify chess pieces in various angles