Unconstrained handwritten Devanagari character recognition using convolutional neural networks

Abstract

In this paper, we introduce a novel offline strategy for recognition of online handwritten Devanagari characters entered in an unconstrained manner. Unlike the previous approaches based on standard classifiers -SVM, HMM, ANN and trained on statistical, structural or spectral features, our method, based on CNN, allows writers to enter characters in any number or order of strokes and is also robust to certain amount of overwriting. The CNN architecture supports an increased set of 42 Devanagari character classes. Experiments with 10 different configurations of CNN and for both Exponential Decay and Inverse Scale Annealing approaches to convergence, show highly promising results. In a further improvement, the final layer neuron outputs of top 3 configurations are averaged and used to make the classification decision, achieving an accuracy of 99.82% on the train data and 98.19% on the test data. This marks an improvement of 0.2% and 5.81%, for the train and test set respectively, over the existing state-of-the-art in unconstrained input. The data used for building the system is obtained from different parts of Devanagari writing states in India, in the form of isolated words. Character level data is extracted from the collected words using a hybrid approach and covers all possible variations owing to the different writing styles and varied parent word structures.

Leaf classification utilizing a convolutional neural network

“…This beat the previous record of 16.7% and significantly improved upon a random guess which is less than 0.005% accuracy. Mehrotra et al [4] implemented a classifier for Devanagari character recognition which obtained 98.19% accuracy. Simard et al [5] trained a classifier for the MNIST handwriting recognition training set that performed with 99.6% accuracy.…”

“…The cost function for the softmax classifier is defined in (6) [6]. The parameters for this equation are the same for (4).…”

Recognition of Online Handwritten Bangla and Devanagari Basic Characters: A Transfer Learning Approach

“…Keeping the above facts in mind, many researchers are now trying to apply transfer learning approach where a pre-trained model, with predefined weights and biases, can be used to train models on different datasets. This in turn saves training time, minimizing implementation complexity associated with initializing weights and biases for layers in DNN models [27]. Chatterjee et al [24] have obtained better recognition accuracy for isolated Bangla-Lekho characters in fewer epochs using ResNet50 through transfer learning approach in comparison to the experiment [25] performed on same dataset.…”

“…Mehrotra et al have introduced an offline strategy to recognize online handwritten Devanagari characters using CNN [27]. Therefore, it can be said that till now so many research attempts have been made using deep learning based approaches in handwriting recognition.…”

Improved deep convolutional neural network for online handwritten Chinese character recognition using domain-specific knowledge

“…By incorporating the path signature feature, DeepCNet produced the best test error rate of 3.58% [13] on CASIA-OLHWDB1.1, which is markedly better than the result of 5.61% from MCDNN [11] or 5.15% from DLQDF [3]. DCNNs have also drawn attention for the applications in the recognition of other Asian character sets such as Hangul [15] and Devanagari [16].…”