Exploiting features from triangle geometry for digit recognition

Triangle is a basic geometry. There are six types of triangle, but scalene triangle was chosen to be used in this research which is based on coordinates of corners generated by our proposed algorithm. In this paper, nine features are proposed. Six of the features were derived from coordinates and sides of triangle whereas three others are angle of corners. After features are identified, the image is zoned into 25 zones. The zoning processes are based on Cartesian plan, Vertical and Horizontal zones. From the zoning process, nine features are expanded to become 225 features. The features proposed are used on HODA, MNIST, IFHCDB and BANGLA datasets. Experiments are conducted using supervised learning which are Support Vector Machine (SVM) and Multi-layer Perceptron (MLP). Results from the experiments are evaluated with different Cost (c) for the SVM and Learning Rate (LR) for the MLP. Then, the result is compared to the state of the art researches in this area.