Design of a Very Compact CNN Classifier for Online Handwritten Chinese Character Recognition Using DropWeight and Global Pooling

Currently, owing to the ubiquity of mobile devices, online handwritten Chinese character recognition (HCCR) has become one of the suitable choice for feeding input to cell phones and tablet devices. Over the past few years, larger and deeper convolutional neural networks (CNNs) have extensively been employed for improving character recognition performance. However, its substantial storage requirement is a significant obstacle in deploying such networks into portable electronic devices. To circumvent this problem, we use a novel technique called DropWeight for pruning redundant connections in the CNN architecture. It is revealed that the method not only treats streamlined architectures such as AlexNet and VGGNet well but exhibits remarkable performance for deep residual network and inception network. We also demonstrate that global pooling is a better choice for building very compact online HCCR systems. Experiments were performed on the ICDAR-2013 online HCCR competition dataset using our proposed network, and it is found that the proposed approach requires only 0.57 MB for storage, whereas state-of-the-art CNN-based methods require up to 135 MB; meanwhile the performance is decreased only by 0.91%.

I. Introduction

Over the last five decades [1], [2], handwritten Chinese character recognition (HCCR) has attracted considerable attention from researchers and has extensively been studied owing to the large number of character classes, similarity between characters, and variation in writing style. Handwriting recognition can broadly be categorized into online and offline handwriting recognition. The main motivation of this paper is to deal with the problem of storage capacity for online HCCR by using some novel techniques. In contrast to offline HCCR, in which gray-scale images are analyzed and classified into different groups, for online HCCR, pen trajectories are the main source of information to recognize different characters [3]. Moreover, online HCCR finds numerous applications in pen input devices, personal digital assistants, smart phones, touch-screen devices, etc.