Handwritten Numbers Recognition, Detection and Generation Based Convolutional Neural Network

Abstract

Convolutional neural network has been widely used in the field of image recognition. It has achieved significantly results in image classification and object detection. Based on those results, the paper discussed the ability of different network structures to recognize handwritten numbers, the effects of fine tuning on the model, and the ability to generate and detect numbers by computer. This paper discussed the structure of CNN and LSTM and compared them in recognition results. We also use the parameters in classifying MNIST to initialize our recognition task. This paper used Faster R-CNN to detect the position of numbers. At last, we used WGAN to discuss the ability that computers generate handwritten numbers.