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

Abstract

Detecting incidental scene text is a challenging task because of multi-orientation, perspective distortion, and variation of text size, color and scale. 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. First, we use quadrilateral sliding windows in several specific intermediate convolutional layers to roughly recall the text with higher overlapping area and then a shared Monte-Carlo method is proposed for fast and accurate computing of the polygonal areas. After that, we designed a sequential protocol for relative regression which can exactly predict text with compact quadrangle. Moreover, a auxiliary smooth Ln loss is also proposed for further regressing the position of text, which has better overall performance than L2 loss and smooth L1 loss in terms of robustness and stability. The effectiveness of our approach is evaluated on a public word-level, multi-oriented scene text database, ICDAR 2015 Robust Reading Competition Challenge 4 "Incidental scene text localization". The performance of our method is evaluated by using F-measure and found to be 70.64%, outperforming the existing state-of-the-art method with F-measure 63.76%.

8 Pages, 7 figures. Accepted to appear in CVPR 2017

References in corpus (5)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Robust Text Detection in Natural Scene Images
• Rich feature hierarchies for accurate object detection and semantic segmentation
• Multi-Oriented Text Detection with Fully Convolutional Networks
• Incidental Scene Text Understanding: Recent Progresses on ICDAR 2015 Robust Reading Competition Challenge 4

Cited by in corpus (25)

• R2CNN: Rotational Region CNN for Orientation Robust Scene Text Detection
• Scene Text Detection and Recognition: The Deep Learning Era
• Single Shot Text Detector with Regional Attention
• Deep Direct Regression for Multi-Oriented Scene Text Detection
• Mask TextSpotter: An End-to-End Trainable Neural Network for Spotting Text with Arbitrary Shapes
• WordSup: Exploiting Word Annotations for Character based Text Detection
• Scene Text Synthesis for Efficient and Effective Deep Network Training
• TextSnake: A Flexible Representation for Detecting Text of Arbitrary Shapes
• Faster RER-CNN: application to the detection of vehicles in aerial images
• Pixel-Anchor: A Fast Oriented Scene Text Detector with Combined Networks
• Sliding Line Point Regression for Shape Robust Scene Text Detection
• Fused Text Segmentation Networks for Multi-oriented Scene Text Detection
• A pooling based scene text proposal technique for scene text reading in the wild
• TextMountain: Accurate Scene Text Detection via Instance Segmentation
• Rotation-Sensitive Regression for Oriented Scene Text Detection
• An Anchor-Free Region Proposal Network for Faster R-CNN based Text Detection Approaches
• Arbitrary Shape Scene Text Detection with Adaptive Text Region Representation
• Geometry Normalization Networks for Accurate Scene Text Detection
• A Survey on Deep Domain Adaptation and Tiny Object Detection Challenges, Techniques and Datasets
• ArbiText: Arbitrary-Oriented Text Detection in Unconstrained Scene
• DeRPN: Taking a further step toward more general object detection
• All You Need Is Boundary: Toward Arbitrary-Shaped Text Spotting
• TextNet: Irregular Text Reading from Images with an End-to-End Trainable Network
• STELA: A Real-Time Scene Text Detector with Learned Anchor
• Accurate Scene Text Detection through Border Semantics Awareness and Bootstrapping