A fast handwritten numeral recognition framework based on peak densities

In this paper, we present a novel framework for handwritten numeral recognition. Considering unconstrained handwritten numerals as numeral feature vectors in the corresponding numeral vector space, we commence by reducing the coordinate dimensionalityof vector space by employing Spectral Regression Discriminant Analysis (SRDA). We then calculate the local density for all numeral classes. For each class, we consider numeral points with local peak densities and large distance from points with higher peak densities as the numeral centers. For the inference tasks, we calculate the average similarity between one testing numeral sample and numeral centers of each digit class. The largest average similarity with numeral centers of one digit class implies that the numeral sample is categorized into this class. In order to validate our framework, experiments with two worldwide standard data sets USPS and MNIST are performed. The experimental results show that the proposed approach achieves high performances in efficiency and robustness for big data analysis.