Script Identification Using Bag-of-Words with Entropy-weighted Patches

The increasing interest in scene text reading in multilingual environments raises the need to recognize and distinguish between different writing systems. In this paper, we propose a novel method for script identification using convolutional features for the traditional bag-of-words model in a combination with weighting by means of intra-cluster information entropy. This approach exploits the expressive representation of convolutional neural networks, which have displayed outstanding performance in many text analysis and recognition tasks in recent years, discriminative power of script-characteristic features, and generalization abilities of bag-of-words model. The proposed method is evaluated on two public benchmark datasets for script identification. The experiments demonstrate that our method outperforms the baseline and yields competitive results.