Precise recognition algorithm for handwritten digit characters based on low-dimensional features
一种基于低维特征的高精度手写数字识别算法

The contour skeleton feature of digital character was proposed. A method based on this feature and the big gridding feature for the recognition of off-line handwritten digits was also developed. The feature vectors extracted were to be recognized and eliminated gradually by making use of the improved two-stage AdaBoost neural network. First stage, the categorizer based on big gridding feature conducted general assortment to eliminate most of negative samples and let almost all the positive samples pass. Second stage, the categorizer based on contour skeleton feature conducted further sorting for the positive samples from the 1st stage. Simulation result indicates that the proposed method has improvement in recognition speed and accuracy rate.