Real-Time Scene Text Detection With Differentiable Binarization and Adaptive Scale Fusion

Liao, Minghui; Zou, Zhisheng; Wan, Zhaoyi; Yao, Cong; Bai, Xiang

doi:10.1109/tpami.2022.3155612

Cited by 208 publications

(87 citation statements)

References 61 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…On the ICDAR2015 dataset, compared to the base model, ADFM improved P by 1.4% to 90.2%, R by 6.4% and F by 4.4%, and performed better than most of the more advanced algorithms in recent years. Compared with the test results of DBNet trained with ResNet-18, the P improved by 3.4%, R by 0.3% and F by 1.8%; compared with the test results of the more recent algorithm DBNet++ [ 36 ] trained on the ResNet-18 backbone, the P is improved by 0.1%, the R is improved by 1.5% and the F is improved by 1%. The results of the comparison with other methods are shown in Table 2 .…”

Section: Methodsmentioning

confidence: 97%

Attention-Based Scene Text Detection on Dual Feature Fusion

Silamu

Wang

et al. 2022

Sensors

View full text Add to dashboard Cite

The segmentation-based scene text detection algorithm has advantages in scene text detection scenarios with arbitrary shape and extreme aspect ratio, depending on its pixel-level description and fine post-processing. However, the insufficient use of semantic and spatial information in the network limits the classification and positioning capabilities of the network. Existing scene text detection methods have the problem of losing important feature information in the process of extracting features from each network layer. To solve this problem, the Attention-based Dual Feature Fusion Model (ADFM) is proposed. The Bi-directional Feature Fusion Pyramid Module (BFM) first adds stronger semantic information to the higher-resolution feature maps through a top-down process and then reduces the aliasing effects generated by the previous process through a bottom-up process to enhance the representation of multi-scale text semantic information. Meanwhile, a position-sensitive Spatial Attention Module (SAM) is introduced in the intermediate process of two-stage feature fusion. It focuses on the one feature map with the highest resolution and strongest semantic features generated in the top-down process and weighs the spatial position weight by the relevance of text features, thus improving the sensitivity of the text detection network to text regions. The effectiveness of each module of ADFM was verified by ablation experiments and the model was compared with recent scene text detection methods on several publicly available datasets.

show abstract

Section: Methodsmentioning

confidence: 97%

Attention-Based Scene Text Detection on Dual Feature Fusion

Silamu

Wang

et al. 2022

Sensors

View full text Add to dashboard Cite

show abstract

“…Table 4 shows the quantitative comparison on the HSText-1000 dataset, we compare our method with some state-of-the-art methods. 1,2,17,20,[33][34][35] Through the experimental comparison, we can find that our method can achieve the best detection performance on the HSText-1000 dataset, which is 10.3% higher than CTPN, 1 43.5% higher than EAST, 2 25.4% higher than CRAFT, 33 31.3% higher than shape robust text detection with progressive scale expansion network (PSE-Net), 34 10.6% higher than ContourNet, 17 3.7% higher than Pan++, 35 and 1.6% higher than DBNet++. 20 This proves that the method proposed in this paper is more competitive than other recent methods for text detection in haze scenes.…”

Section: Comparison With Other Methodsmentioning

confidence: 99%

EWST: an extreme weather scene text detector with dehazing and localization refinement

et al. 2023

View full text Add to dashboard Cite

In real life, the emergence of haze brings great inconvenience and does harm to traffic and pedestrian safety, whereas previous studies paid less attention to text detection in haze scenes. In this experiment, we found that the candidates obtained by the general non-maximum suppression (NMS) method or the soft-NMS method are difficult to precisely match the ground truth, and the incomplete feature extraction will affect the final performance. In this work, a haze scene text detection framework is skillfully designed. An optimized NMS and an optimized long short-term memory for spatial feature extraction and temporal feature extraction are proposed to improve the text detection performance. In addition, a hazing scene text dataset (named HSText-1000) and a hybrid scenario text dataset (named MHSText-4600) have been built in our work for evaluating the performance conveniently, which have been released and can be downloaded from https://github.com/lyy0117/lyy. Experimental results illustrate that our method is superior to some state-of-the-art methods in the hazing scene and the hybrid scene. Meanwhile, we achieved competitive results in nonhaze's public dataset (ICDAR 2013), which means that our method has satisfactory adaptability. We will release code to facilitate community research.

show abstract

“…Other detection methods are usually based on image segmentation, for example, CRAFT [13] proposed detected text area by exploring each character and the affinity between characters., Liao et al. [14] proposed a Differentiable Binarization (DB) module that integrates the binarization process and an efficient Adaptive Scale Fusion (ASF) module to improve the scale robustness by fusing features of different scales adaptively. Text recognition task includes Connectionist Temporal Classification (CTC) based methods [15] and encoder‐decoder methods [16].…”

Section: Related Workmentioning

confidence: 99%

“…In the detection task of the deep learning era, some methods are focused on the bounding boxes: Deng et al [11] detected multi-oriented text with corner-based region proposals by the precise setting of the anchor, and Zhong et al [12] proposed a novel progressive region prediction network (PRPN) with directional pooling for predicting text regions. Other detection methods are usually based on image segmentation, for example, CRAFT [13] proposed detected text area by exploring each character and the affinity between characters., Liao et al [14] proposed a Differentiable Binarization (DB) module that integrates the binarization process and an efficient Adaptive Scale Fusion (ASF) module to improve the scale robustness by fusing features of different scales adaptively. Text recognition task includes Connectionist Temporal Classification (CTC) based methods [15] and encoder-decoder methods [16].…”

Section: Text Detection and Recognitionmentioning

confidence: 99%

Method of railway shunting operation sheet information extraction guided by table header

Zhang

Huang

Zhang

et al. 2022

IET Intelligent Trans Sys

View full text Add to dashboard Cite

The railway shunting operation sheet is the written plan of the shunting operation. In the current shunting operation preparation process, signalmen must input the plans manually for subsequent execution, which is inefficient and at risk of misoperation. Here, an information extraction method is proposed to convert shunting operation sheet images to readable instructions for Continuous Train Control (CTC) systems. First, the shunting operation sheet images are preprocessed to facilitate subsequent text recognition based on the Paddle OCR model. Second, the table is located and reconstructed with the results of recognized contents and coordinates of texts. Then, the results of the table reconstruction were optimized based on the tabulation rules of shunting operation. Finally, tabular texts containing structural and attribute information are transcribed into eXtensible Markup Language (XML) format, which can be read automatically by CTC systems based on the C programming language. The proposed approach was evaluated on real‐world samples, and the results showed that the comprehensive accuracy of information extraction was 0.85. Furthermore, a pre‐processing combination strategy was established to improve the accuracy and generalizability of the proposed method.

show abstract

Real-Time Scene Text Detection With Differentiable Binarization and Adaptive Scale Fusion

Cited by 208 publications

References 61 publications

Attention-Based Scene Text Detection on Dual Feature Fusion

Attention-Based Scene Text Detection on Dual Feature Fusion

EWST: an extreme weather scene text detector with dehazing and localization refinement

Method of railway shunting operation sheet information extraction guided by table header

Contact Info

Product

Resources

About