基于改进的支持向量机的手写体汉字识别

Handwritten Chinese Character Recognition Based on the Improved Support Vector Machine

逐次超松弛迭代法算法是一种具体的SVM算法,在SOR算法中松弛因子采取固定数值时,在许多情况下收敛速度较慢。文中提出通过引入具有"先验知识"的神经网络,对逐次超松弛迭代法中的松弛因子进行控制,以提高逐次超松弛迭代法的收敛速度。实验结果表明,该模型实现的逐次超松弛迭代法能够提高其收敛速度。在手写体汉字的识别实验中,该改进算法可以减少支持向量机的训练时间。

Successive over relaxation iteration algorithm is one of classical SVM algorithms. In SOR algorithm, if the relaxation factor is a constant, convergent speed will be relatively slow in many conditions. The paper introduces a neural network with ＂transcendental knowledge＂ to control relaxation factor to get faster convergent speed. The experiment result shows the SOR iteration method using this model can effectively advance the convergent speed. In the handwritten Chinese character recognition test, the improved algorithm cuts down the training time of SVM.