Building Fast and Compact Convolutional Neural Networks for Offline Handwritten Chinese Character Recognition

TL;DR

This paper introduces a novel CNN optimization method that significantly reduces computational cost and storage for offline handwritten Chinese character recognition, enabling fast and compact deployment on portable devices.

Contribution

The authors propose GSLRE and ADW techniques to create a nine-layer CNN that is nine times faster and 18 times smaller with minimal accuracy loss, outperforming existing models.

Findings

Network size reduced to 1/18 with only 0.21% accuracy drop.

Recognition speed achieved 9.7 ms per character on CPU.

Model is approximately 30 times faster and 10 times more cost-efficient than state-of-the-art CNNs.

Abstract

Like other problems in computer vision, offline handwritten Chinese character recognition (HCCR) has achieved impressive results using convolutional neural network (CNN)-based methods. However, larger and deeper networks are needed to deliver state-of-the-art results in this domain. Such networks intuitively appear to incur high computational cost, and require the storage of a large number of parameters, which renders them unfeasible for deployment in portable devices. To solve this problem, we propose a Global Supervised Low-rank Expansion (GSLRE) method and an Adaptive Drop-weight (ADW) technique to solve the problems of speed and storage capacity. We design a nine-layer CNN for HCCR consisting of 3,755 classes, and devise an algorithm that can reduce the networks computational cost by nine times and compress the network to 1/18 of the original size of the baseline model, with only a 0.21% drop in accuracy. In tests, the proposed algorithm surpassed the best single-network performance reported thus far in the literature while requiring only 2.3 MB for storage. Furthermore, when integrated with our effective forward implementation, the recognition of an offline character image took only 9.7 ms on a CPU. Compared with the state-of-the-art CNN model for HCCR, our approach is approximately 30 times faster, yet 10 times more cost efficient.