R2CNN: Rotational Region CNN for Orientation Robust Scene Text Detection

By Yingying Jiang, Xiangyu Zhu, Xiaobing Wang, Shuli Yang, Hua Wang, Pei Fu, Zhenbo Luo. arXiv.org, 2017

566 citations recorded by the source corpus.

In this paper, we propose a novel method called Rotational Region CNN (R2CNN) for detecting arbitrary-oriented texts in natural scene images. The framework is based on Faster R-CNN [1] architecture. First, we use the Region Proposal Network (RPN) to generate axis-aligned bounding boxes that enclose the texts with different orientations. Second, for each axis-aligned text box proposed by RPN, we extract its pooled features with different pooled sizes and the concatenated features are used to simultaneously predict the text/non-text score, axis-aligned box and inclined minimum area box. At last, we use an inclined non-maximum suppression to get the detection results. Our approach achieves competitive results on text detection benchmarks: ICDAR 2015 and ICDAR 2013.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/16da4d6503e17f8597602437358461c252244bf7

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Junchi Yan
2. Xue Yang
3. Xiang Bai
4. Luc Van Gool
5. Dacheng Tao
6. Junwei Han
7. Chunhua Shen
8. Qi Tian

In Media Technology

1. Guisong Xia
2. Cong Yao
3. Xiang Bai
4. Minghui Liao
5. Gong Cheng
6. Xin He
7. Zhenwei Shi
8. Junwei Han

Citations in the corpus

• MMRotate: A Rotated Object Detection Benchmark using PyTorch (2022, methodology): “R2CNN [12] simultaneously predicts the axis-aligned and inclined boxes by adding an inclined box branch and uses an inclined NMS to obtain the detection results.”
• R3Det: Refined Single-Stage Detector with Feature Refinement for Rotating Object (2019, methodology): “As for scene text detection, there are many efficient rotation detection methods, such as two-stage methods: R 2 CNN [15], RRPN [ ?”
• Scene Text Detection and Recognition: The Deep Learning Era (2018, methodology): “In this stage, scene text detection algorithms are designed by modifying the region proposal and bounding box regression modules of general detectors to localize text instances directly (Dai et al. 2017; He et al. 2017c; Jiang et al. 2017; Liao et al. 2017, 2018a; Liu and Jin 2017; Shi et al. 2017a;”
• Learning RoI Transformer for Oriented Object Detection in Aerial Images (2019, methodology): “The HRoI based methods [15, 37] usually use a normal RoI Warping to extract feature from a HRoI, and regress position off-sets relative to the ground truths.”
• Learning High-Precision Bounding Box for Rotated Object Detection via Kullback-Leibler Divergence (2021, methodology): “The latter is based on the length and width of the anchor (wa, ha) to adjust the gradient instead of the target object (wt, ht), which is almost ineffective for those detectors [3, 13, 15, 27, 28, 32, 33] that use horizontal anchors for rotation detection.”
• Dense Label Encoding for Boundary Discontinuity Free Rotation Detection (2020, methodology): “Many rotation detectors (including those for aerial image [46, 15, 47, 2, 6, 51, 45, 30, 48, 49, 50], scene text [56, 23, 14, 26, 19, 18] and face [34, 13, 33]) are derived based on the vanilla detection algorithms [10, 32, 31, 5, 20, 21].”
• Automatic Ship Detection in Remote Sensing Images from Google Earth of Complex Scenes Based on Multiscale Rotation Dense Feature Pyramid Networks (2018, methodology): “Compared with other rotation region detection methods such as RRPN [27] and R2CNN [26], our framework is more suitable for ship detection tasks, and has achieved state-of-the-art performance.”
• Sliding Line Point Regression for Shape Robust Scene Text Detection (2018, methodology): “[11] firstly generated axis-aligned bounding boxes and then regressed the text quadrangle.”
• TextBoxes++: A Single-Shot Oriented Scene Text Detector (2018, methodology): “To ease this problem, we propose to use another representation proposed in [44] for a ground truth rotated rectangle Gr: Gr = (x̃1, ỹ r 1, x̃ r 2, ỹ r 2, h̃ ), where (x̃1, ỹ r 1) and (x̃ r 2, ỹ r 2) are the first and second vertex of the ground truth (i.”
• Enhancing Geometric Factors in Model Learning and Inference for Object Detection and Instance Segmentation (2020, methodology): “There are other detectors such as RRPN [27] and R 2 CNN [46] adding rotation angle regression to detect arbitrary orientated objects for remote sensing detection and scene text detection.”
• Shape Robust Text Detection With Progressive Scale Expansion Network (2018, methodology): “For the regression-based approaches [36, 43, 32, 16, 42, 23, 11, 13, 27], the text targets are usually represented in the forms of rectangles or quadrangles with certain orientations.”
• SCRDet++: Detecting Small, Cluttered and Rotated Objects via Instance-Level Feature Denoising and Rotation Loss Smoothing (2020, background): “[16], [17], [18], in an out of box manner.”
• Rethinking Rotated Object Detection with Gaussian Wasserstein Distance Loss (2021, background): “Arbitrary-oriented objects are ubiquitous for detection across visual datasets, such as aerial images [65, 2, 8, 69], scene text [82, 31, 20, 35, 27], faces [47] and 3D objects [78], retail scenes [5, 41], etc.”
• Object Detection in 20 Years: A Survey (2019, background): “…to this problem is to introduce additional parameters in anchor boxes and RoI pooling layer that are associated with rotation and perspective changes [351–353 To improve densely arranged text detection : The segmentation-based approach shows more advantages in detecting densely arranged texts.”