基于卷积神经网络的手写数字识别方法研究

Research on handwritten digits recognition method based on Convolutional Neural Network

手写数字识别是计算机视觉的一项典型应用,其成果可广泛应用于邮政编码识别、统计报表识别、考试成绩判定等领域。针对单幅图像中多个手写数字的自动分割及识别问题,文中采用自适应二值化方法实现手写数字与背景的分割,利用基于方向投影的改进算法将各个数字依次进行分割提取,通过手写Minist训练数据集对卷积神经网络的LeNet-5模型进行训练,利用Tensorflow实现了单幅图像内多个手写数字的分割与识别。实验结果表明,文中方法具有较高的可靠性,训练后的模型对新的手写数字平均识别率在92%以上,达到了预期的效果。

Handwritten digits recognition is a typical application of computer vision which can be applied to the fields of postal code recognition,statistical report recognition,test result determination and so on.Aiming at the problem of automatic segmentation and recognition of multiple handwritten digits in a single image,this paper uses an adaptive binarization method to achieve the segmentation of handwritten digits from the background,and uses an improved algorithm based on directional projection to segment and extract each digit in turn,through handwritten Minist training data realized the training of LeNet-5model of the Convolutional Neural Network,and uses Tensorflow to realize the segmentation and recognition of multiple handwritten digits in a single image. According to the experimental results,the method in this paper has high reliability. The trained model has an average recognition rate of over 92%for new handwritten digits which achieving the expected recognition result.