Handwriting digits recognition is an important research subject in character recognition. The difficulties of handwriting digits recognition are due to its various anamorphosis. At present, the digits recognitions of different letterforms, especially the offline handwriting digits recognition, are still under development, and the recognition effect is not ideal. Therefore, it is still a very important research direction to study simple and high-efficient handwriting digits recognition.The thesis probes into the key issue of handwriting digits recognition—feature extraction and feature selection. The main work of the thesis includes the following aspects:1. Based on the researches on the features of several handwriting digits, the thesis extracts the structures and statistic features of seven kinds of handwriting digits, i.e. outline feature, stroke density feature, wide grid feature, barycenter and barycenter distance feature, the first black point position feature, project feature, and Fourier switch feature.2. From different feature selection methods, this thesis adopts three methods—inner and outer analogy, K-W checking and entropy function—to select the features.3. This thesis analyzes the feature dimension decrease issue of the handwriting digits through a lot of experiments.4. This thesis establishes a handwriting digit recognition system based on BP neural network. The original features and selected features both have good systematic performance after checked through BP neural network, which proves the above mentioned method feasible. |