Detecting text in natural scenes with stroke width transform

By Boris Epshtein, Eyal Ofek, Yonatan Wexler. 2010

1,668 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology

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

In Media Technology

Citations in the corpus

Synthetic Data and Artificial Neural Networks for Natural Scene Text Recognition (2014, methodology): “Previous work uses sliding window methods [26] or region grouping methods [5, 6, 19] very successfully for this.”
Reading Text in the Wild with Convolutional Neural Networks (2014, methodology): “These focus solely on text detection (Anthimopoulos et al. 2013; Chen et al. 2011; Epshtein et al. 2010; Gomez and Karatzas 2013, 2014; Huang et al. 2014; Yi and Tian 2011; Yin et al. 2013), text recognition (Almazán et al.”
End-to-End Text Recognition with Hybrid HMM Maxout Models (2013, methodology): “Alternative methods that cleverly exploit the nature of text such as Maximally Stable Extremal Regions (MSERs) (Matas et al., 2004) and Stroke Width Transform (Epshtein et al., 2010) generally have lower accuracy but are fast to compute.”
Real-time scene text localization and recognition (2012, methodology): “in [5] converts an input image to a greyscale space and uses Canny detector [1] to find edges.”
TextProposals: A text-specific selective search algorithm for word spotting in the wild (2016, methodology): “model [18, 26, 33, 25], or because they learn an optimal grouping strategy from examples of well-segmented character groupings [29, 34].”
Text Localization in Real-World Images Using Efficiently Pruned Exhaustive Search (2011, methodology): “The rules require that height ratio, centroid angle and region distance normalized by region width fall within a given interval obtained in a training stage (similar binary rules have been used in many previous works [5], [6], [7], [4]).”
Robust Seed Localization and Growing with Deep Convolutional Features for Scene Text Detection (2015, methodology): “Many existing methods are devoted to the ﬁrst stage, which can be further categorized into two groups: the exhaustive search based methods [1, 10, 16, 18] and the selective search based methods [2, 12, 13, 6, 14, 21, 3, 9].”
ASTER: An Attentional Scene Text Recognizer with Flexible Rectification (2019, methodology): “Representative proposal extraction methods include Maximally Stable Extremal Regions (MSER) [43], [46], Extremal Regions (ER) [47], and Stroke Width Transform (SWT) [14], [65], [66].”
Mask TextSpotter: An End-to-End Trainable Neural Network for Spotting Text with Arbitrary Shapes (2018, background): “A large number of methods have been proposed to detect scene text [7, 36, 37, 50, 19, 23, 54, 21, 47, 54, 56, 30, 52, 55, 34, 15, 48, 43, 57, 16, 35, 31].”
$L_0$ -Regularized Intensity and Gradient Prior for Deblurring Text Images and Beyond (2017, methodology): “While this algorithm achieves state-of-the-art deblurring results, the kernel estimation process is complicated and the performance depends on whether or not the stroke width transform (SWT) [12] separates an image into text and non-text regions well.”
A robust arbitrary text detection system for natural scene images (2014, methodology): “Inspired by the Stroke Width Transform (SWT) Epshtein et al., 2010 which selects text components based on the gradient direction of each pixel, we extend the same idea in a novel way to propose three different symmetry properties to separate text pixels from the non-text pixels in edge images of the”
Detecting Curve Text in the Wild: New Dataset and New Solution (2017, neutral): “hout post-processing. Algorithm Whole set Non-curve subset Curve subset S (FPS) R (%) P (%) H (%) R (%) P (%) H (%) R (%) P (%) H (%) Seglink [27] 40.0 42.3 40.8 48.4 38.3 42.8 19.4 9.9 13.2 10.7 SWT [6] 9.0 20.7 12.5 5.8 13.4 8.1 6.4 7.0 6.7 - CTPN [32] 53.8 60.4 56.9 59.4 54.3 56.7 37.7 34.1 35.8 ”
Total-Text: A Comprehensive Dataset for Scene Text Detection and Recognition (2017, neutral): “[15] and Neumann and Matas [16].”
Character Region Awareness for Text Detection (2019, methodology): “The major trend in scene text detection before the emergence of deep learning was bottom-up, where hand-crafted features were mostly used – such as MSER [27] or SWT [5]– as a basic component.”
Multi-oriented Text Detection with Fully Convolutional Networks (2016, methodology): “By considering both of the two level cues at the same time, our approach has two advantages compared to component based methods [16, 4, 6].”