一种基于低维特征的高精度手写数字识别算法

Precise recognition algorithm for handwritten digit characters based on low-dimensional features

提出了数字字符的轮廓骨架特征,并将这一特征与粗网格特征相结合对脱机手写体数字进行识别。获取特征向量后,利用改进的基于两级级联结构的AdaBoost神经网络进行逐层淘汰识别。第一级首先使用基于粗网格特征的分类器进行粗分类,淘汰大部分负样本,而使几乎所有的正样本通过。第二级由基于轮廓骨架特征的分类器对通过第一级的样本进一步淘汰识别。仿真结果表明,该办法在识别速度与识别率方面都有较大幅度的改进。

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 1 st stage. Simulation result indicates that the proposed method has improvement in recognition speed and accuracy rate.