Character Region Awareness for Text Detection

Abstract

Scene text detection methods based on neural networks have emerged recently and have shown promising results. Previous methods trained with rigid word-level bounding boxes exhibit limitations in representing the text region in an arbitrary shape. In this paper, we propose a new scene text detection method to effectively detect text area by exploring each character and affinity between characters. To overcome the lack of individual character level annotations, our proposed framework exploits both the given characterlevel annotations for synthetic images and the estimated character-level ground-truths for real images acquired by the learned interim model. In order to estimate affinity between characters, the network is trained with the newly proposed representation for affinity. Extensive experiments on six benchmarks, including the TotalText and CTW-1500 datasets which contain highly curved texts in natural images, demonstrate that our character-level text detection significantly outperforms the state-of-the-art detectors. According to the results, our proposed method guarantees high flexibility in detecting complicated scene text images, such as arbitrarily-oriented, curved, or deformed texts.

ICDAR2019 Robust Reading Challenge on Arbitrary-Shaped Text - RRC-ArT

“…This method also has three major modules: detection, rectification, and recognition. Specifically, it adopts CRAFT [13] as its text detector, Thin-Plate-Spline (TPS) based Spatial Transformer Network as its image normalizer, and a BiLSTM with attention as its text recognizer. Figure 5a shows some successful examples of the said method, it appears that the method is robust against curved text instances on both the Chinese and Latin scripts.…”

Scene Text Detection with Polygon Offsetting and Border Augmentation

“…Dataset and F-Measure results FPS ICDAR2015 ICDAR2017 ICDAR2019 Total-Text CTPN [1] 60.9 ---7.5 EAST [4] 76.4 ---17.1 SegLink [8] 75.0 ---12.2 TextBoxes++ [7] 81.7 ---13.2 R2CNN [3] 82.5 ----PixelLink [9] 83.7 ----TextSnake [21] 82.6 --78. 4 12.7 PSENet [22] 87.1 72.1 -80.9 9.6 SPCNET [20] 87.2 70.0 -82.9 -Pixel-Anchor [33] 87.7 68.1 ---PMTD [34] 89.3 78.5 82.5 --CRAFT [35] 86.9 73.9 70.9 83.6 11.2 LOMO [19] 86…”

Multi-Lingual Text Localization via Language-Specific Convolutional Neural Networks

“…In this work, we used three datasets widely employed to design and evaluate text localization and recognition methods, the ICDAR 2015 [4], MLT 2017 [5], and MLT 2019 [2] datasets, which are described in this section. Figure 3 illustrates examples from these datasets.…”

“…However, when we analyze the F-measure per language, we observe a large difference in performance that reaches 36.97%, considering the Latin and Bangla languages. Similarly, CRAFT method [2] presented an overall F-measure of 74.03% and a difference in performance, considering the best and the worst results for each language separately, of 41.66%. The same phenomenon was observed for other approaches, such as the PixelLink network [3].…”