Deep Matching Prior Network: Toward Tighter Multi-oriented Text Detection

Retrospective research has only focused on using rectangular bounding box or horizontal sliding window to localize text, which may result in redundant background noise, unnecessary overlap or even information loss. To address these issues, we propose a new Convolutional Neural Networks (CNNs) based method, named Deep Matching Prior Network (DMPNet), to detect text with tighter quadrangle.

quadrangle 四边形

• firstly, roughly recalling text with quadrilateral sliding window
• then, using a shared Monte-Carlo method for fast and accurate computing of polygonal areas;
• finely localizing text with quadrangle and design a Smooth Ln loss for
  moderately adjusting the predicted bounding box.

设计多几个sliding window

改进计算overlap的方式

回归十个参数

In future, we will explore using shape-adaptive sliding windows toward tighter scene text detection.