Beyond human recognition: A CNN-based framework for handwritten character recognition

Abstract

Because of the various appearance (different writers, writing styles, noise, etc.), the handwritten character recognition is one of the most challenging task in pattern recognition. Through decades of research, the traditional method has reached its limit while the emergence of deep learning provides a new way to break this limit. In this paper, a CNN-based handwritten character recognition framework is proposed. In this framework, proper sample generation, training scheme and CNN network structure are employed according to the properties of handwritten characters. In the experiments, the proposed framework performed even better than human on handwritten digit (MNIST) and Chinese character (CASIA) recognition. The advantage of this framework is proved by these experimental results.

Online and offline handwritten Chinese character recognition: A comprehensive study and new benchmark

“…Moreover, our method also has the lowest memory usage compared with all the other systems, due to the compact representation of directMap and our special convNet structure. For example, the previous best performance achieved by [41] (12th row) is based on the ensemble of 5 networks with memory consumption 950MB, while our model (13th row) is a single network of 23.5MB. In Table 2, the methods from 4th to 16th are all based on convNets, which implies that deep learning based methods are becoming more and more popular for solving offline HCCR problem.…”

“…The ensemble of multiple models is widely used to achieve the highest accuracies on different databases [37,15,43,41]. Model ensemble can be achieved by training the same convNet multiple times with different random initializations and different mini-batches (due to random data shuffling).…”

“…However, deeper network, proper sample distortion, and multi-supervised training [81] are required to guarantee the performance. It is shown in [41] that: without the multi-supervised training, the network is not able to converge during the training. Moreover, to eliminate the influence from different handwriting styles, convNet trained on raw data is usually designed to be very large (e.g.…”

Image Distortion Detection Using Convolutional Neural Network

“…The main model we choose for our distortion recognition system is a convolutional neural network (CNN) [14]. CNNs have been widely used and verified over a variety of image understanding tasks [5,13,15]. The overall structure we use is a 'Y'-shaped CNN that performs distortion classification and detection simultaneously.…”

Advances in next era cloud-empowered computing and techniques

“…The authors describe cloud-based system to share computation resources to reduce redundant computation based on hybrid CNNs (convolutional neural networks) [36,37]. The proposed system combines several trained CNNs through a cloud server, which are used to make different feature recognition based on frames from webcam stream.…”