Deep Neural Network for Handwritten Digital Recognition Based on Attention Mechanism
Download as PDF
Yibo Hao, Jianbo Chen
Handwritten digital recognition is a hot spot in the field of artificial intelligence, which has already played a very important role in the society. A lot of handwritten digital recognition algorithms have been developed in the last few decades. However, most of previous algorithms failed to better extract the semantic and useful local parts from the original image, which is important for the enhancement of signal noise ratio. In this paper, an attention mechanism based neural network is proposed to better extract the salient local information automatically, and simultaneously filter out background or noise in the handwritten digital images. By this way, our method is able to only focus on the important information of the input image, and consequently better understanding the semantic information of the image. Experiments has been conducted on a widely used dataset MNIST, and the results show that our method is able to achieve great performance on the handwritten digital recognition task.
Convolutional neural network, Attention mechanism, Handwritten digital recognition