Scene Text Detection Using Context-Aware Pyramid Feature Extraction
Qishu Jian · 2020
In the past decades, scene text detection has been attached much attention. Many researchers carried out researches on deep learning or neural network. However, the common scene text detection methods are sometimes inadequate because of the various forms and complex characteristics in natural scene images. To overcome those deficiencies, we propose an algorithm combing the Efficient and Accurate Scene Text Detector Model (EAST) with the Context-aware Pyramid Feature Extraction module (CPFE). Built on the base of fully convolutional neural network (FCN), CPFE is integrated to expend receptive field and obtain multi-scale and high-level features from context, which largely overcome the deficiency in scene text detection. Experiments conducted on the ICDAR2015 dataset show that our model outperforms the original EAST model.