PROJECT TITLE :

Scene Text Detection and Segmentation based on Cascaded Convolution Neural Networks - 2017

ABSTRACT:

Scene text detection and segmentation are 2 important and difficult research problems in the sphere of pc vision. This paper proposes a novel methodology for scene text detection and segmentation primarily based on cascaded convolution neural networks (CNNs). In this method, a CNN-primarily based text-aware candidate text region (CTR) extraction model (named detection network, DNet) is designed and trained using each the edges and the entire regions of text, with that coarse CTRs are detected. A CNN-based CTR refinement model (named segmentation network, SNet) is then created to exactly phase the coarse CTRs into text to induce the refined CTRs. With DNet and SNet, a lot of fewer CTRs are extracted than with traditional approaches whereas additional true text regions are kept. The refined CTRs are finally classified using a CNN-based CTR classification model (named classification network, CNet) to urge the ultimate text regions. All of these CNN-primarily based models are modified from VGGNet-16. In depth experiments on three benchmark data sets demonstrate that the proposed technique achieves the state-of-the-art performance and greatly outperforms alternative scene text detection and segmentation approaches.