基于原型生成技术的手写体数字识别

Handwriting digit recognition based on prototype generation technique

为进一步从已有实验方法中提高手写体数字识别识别率,降低数字识别的时间,提出基于原型生成技术的实验方案。该技术包含两部分的处理过程,第一阶段应用自适应共振理论1(ART1)从原有训练集中选择概括全部特性的原型集合的解决方案,第二阶段利用自然演化理论的执行生成最优原型,使目标函数最小化;采用k-NN(k-nearest neighbor)邻近算法进行手写体分类识别。在ART1的基础上,利用自然演化策略的方式改变原型生成技术对MNIST数据库的识别,验证了该技术能够很好权衡识别准确率、分类速度和手写体风格变化的稳定性。

To improve the recognition rate of handwriting digit recognition and reduce the time of digital identification of the existing experimental methods, a prototype generation technique was presented. Prototype generation was approached as a two- stage process. In the first stage, an adaptive resonance theory 1 （ART1） based algorithm was used to select an effective initial solution, while in the second one, the natural evolution strategy was used to generate the best prototypes, and the objective function was minimized. The classification task was performed using the k-nearest neighbor classifier. The prototype generation technology was changed by using natural evolution strategy approach based on ART1 to identify the MNIST database. The pro- posed technique was verified to be able to represent a good trade-off among accuracy, classification speed and robustness to hand- writing style changes.