Robust Text Detection in Natural Scene Images

By Van Pham, Gueesang Lee. Australasian Conference on Artificial Intelligence, 2016

635 citations recorded by the source corpus.

Natural scenes of blurred text images are challenges to current text recognition field. In our paper, a novel method for text detection in natural scene image is suggested using edge detection, maximally stable extremal region (MSER) and tensor voting. Edge detection and MSER methods are combined to find the greatest character candidates from stable areas which are extracted from an input image. These text candidates are used to extract the text line information using tensor voting that creates normal vectors and curve saliency values in characters along the text lines. Therefore, the text line information is used to eliminate non-text areas. Our method is evaluated on the ICDAR2013 datasets and experiment results show that the proposed result is compared to the previous methods.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/266f1172c722cc55e2b0a23b5b479258a071b65f

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. Xiang Bai
2. Cong Yao
3. Chunhua Shen
4. Yu Qiao
5. Luc Van Gool
6. Dacheng Tao
7. Minghui Liao
8. Xiaogang Wang

Citations in the corpus

• Robust Seed Localization and Growing with Deep Convolutional Features for Scene Text Detection (2015, methodology): “Yin et al. [20] exploits MSER to extract character candidates and remove the re-peating components by several heuristic rules.”
• Total-Text: A Comprehensive Dataset for Scene Text Detection and Recognition (2017, background): “Last but not least, we ran [22] and [6] on several images of Total-Text, results can be seen in Fig.”
• TextSnake: A Flexible Representation for Detecting Text of Arbitrary Shapes (2018, background): “Before the era of deep learning, SWT [3] and MSER [24] are two representative algorithms that have influenced a variety of subsequent methods [42, 11].”
• Recognizing Text-Based Traffic Guide Panels with Cascaded Localization Network (2016, neutral): “Most state-of-the-art methods of scene text extraction [1,2,4,5,6,7,8,9,10] comprise two stages, detection to obtain image regions containing text information, and recognition to transform image-based text information into text codes.”
• Arbitrary Shape Scene Text Detection With Adaptive Text Region Representation (2019, methodology): “CC based methods, especially the Maximally Stable Extremal Regions (MSER) based methods [26, 30], get character candidates by extracting CCs.”
• Scene text detection and recognition: recent advances and future trends (2015, methodology): “All the algorithms except for [12, 28, 34, 49] can only deal with horizontal or nearhorizontal texts in natural scenes.”
• Text-Attentional Convolutional Neural Network for Scene Text Detection (2015, methodology): “In [1], [5], [23], and [26], the MSERs detector”
• Towards End-to-End Text Spotting with Convolutional Recurrent Neural Networks (2017, neutral): “Method ICDAR2015 Word-Spotting ICDAR2015 End-to-End ICDAR2011 Word-Spotting SVT Word-Spotting Strong Weak Generic Strong Weak Generic Generic Strong Generic Deep2Text II+ [31] 84.”
• EAST: An Efficient and Accurate Scene Text Detector (2017, background): “features are manually designed [5, 25, 40, 10, 26, 45] to capture the properties of scene text, while in deep learning based methods [3, 13, 11, 12, 7, 48] effective features are directly learned from training data.”
• Detecting Oriented Text in Natural Images by Linking Segments (2017, methodology): “The inference process of SegLink is a single forward pass in the detection network, while the previous methods [25, 28, 30] involve sophisticated rule-based grouping or filtering steps.”
• Reading Scene Text in Deep Convolutional Sequences (2015, background): “This problem includes two sub tasks, namely text detection [16, 34, 35, 15, 24] and text-line/word recognition[18, 1, 17, 4, 33].”
• A Fast Uyghur Text Detector for Complex Background Images (2018, methodology): “The approach has achieved the best performance on ICDAR 2011 [5].”
• Detecting Text in Natural Image with Connectionist Text Proposal Network (2016, methodology): “Current approaches for text detection mostly employ a bottom-up pipeline [28,1,14,32,33].”
• Single Shot Text Detector with Regional Attention (2017, methodology): “Previous works on scene text detection mainly focus on bottom-up approaches, which detect characters or text components from images by using hand-crafted features [27, 33, 10, 31] or sliding window methods [12, 30].”