Complexity and Accuracy Analysis of Convolutional Neural Networks in Scene Text Detection

Machine vision, deep learning and convolutional neural networks have been applied to scene text detection by more and more scholars, whom proposed many excellent methods like Faster R-CNN, Connectionist Text Proposal Network (CTPN) and Arbitrary-Oriented Scene Text Detection via Rotation Proposals (RRPN). This article analyzes the various indicators and detection accuracy of these three methods. On the network architecture, Faster R-CNN is more complex than RRPN and CTPN. In terms of detection accuracy, the F-measure of RRPN is 0.18 higher than Faster R-CNN; the F-measure of CTPN is 0.15 higher than Faster R-CNN. However, these three methods have different emphasis on the detected text in the process of scene text detection. CTPN has better performance for laterally distributed scene text detection; RRPN is superior for scene text detection with tilt angle.