Applications of neural networks in handwritten digit recognition

This paper proposes a handwritten digit recognition system based on a HyperRectangular Composite Neural Network(HRCNN). This system consists of four parts : an input device, preprocessing, feature extraction, and a classifier. The input device is an image, scanner which can get several numerals simultaneously. Preprocessing involves image segmentation and normalization. The Localized Arc Patterns are used to extract the feature vectors from the normalized image. Then the feature vectors are used as the input of the HRCNN and we can obtain the result. The experiment shows a good result and thus we can use this system in a wide range of applications about handwritten digit recognition.