Method for unconstrained text detection in natural scene image

Abstract

Text detection in natural scene images is an important prerequisite for many content-based multimedia understanding applications. The authors present a simple and effective text detection method in natural scene image. Firstly, MSERs are extracted by the V-MSER algorithm from channels of G, H, S, O 1 , and O 2 , as component candidates. Since text is composed of character candidates, the authors design an MRF model to exploit the relationship between characters. Secondly, in order to filter out non-text components, they design a set of two-layers filtering scheme: most of the non-text components can be filtered by the first layer of the filtering scheme; the second layer filtering scheme is an AdaBoost classifier, which is trained by the features of compactness, horizontal variance and vertical variance, and aspect ratio. Then, only four simple features are adopted to generate component pairs. Finally, according to the orientation similarity of the component pairs, component pairs which have roughly the same orientation are merged into text lines. The proposed method is evaluated on two public datasets: ICDAR 2011 and MSRA-TD500. It achieves 82.94 and 75% F-measure, respectively. Especially, the experimental results, on their URMQ_LHASA-TD220 dataset which contains 220 images for multi-orientation and multi-language text lines evaluation, show that the proposed method is general for detecting scene text lines in different languages.

Open Datasets and Tools for Arabic Text Detection and Recognition in News Video Frames

“…This dataset includes 500 images (300 for training and 200 for testing) with horizontal and slant/skewed texts in complex natural scenes (see Figure 2b for examples). The method of Liu et al [27] achieves state-of-the-art performance on this database with an F-score of 75%. This method makes use of the Maximally Stable Extremal Regions (MSER) technique as text candidates extractor as well as a set of heuristic rules and an AdaBoost classifier as a two-stages filtering process.…”

Curve description by histograms of tangent directions

“…The reason we restrict the experiment to MSER detector output is to separate the descriptor performance from the preprocessing stages. This lets us estimate the value of the HOTD descriptor for scene text recognition pipelines that also exploit the MSER detector for character detection [28]. On the Chars74k data set, we also test the classification performance coupled with segmentation algorithms, which is the standard protocol for the natural images subset of Chars74k .…”

Specific category region proposal network for text detection in natural scene

“…Connected region‐based methods suppose that pixels in the text region have similar characteristic in some aspects, such as colour, brightness, texture etc. Stroke width transform[17, 33] and MSER [18–20, 34] are two typical connected region‐based methods. Liu et al [18] proposed a V‐MSER algorithm to extract MSERs from multi‐channel for text detection.…”

“…Stroke width transform[17, 33] and MSER [18–20, 34] are two typical connected region‐based methods. Liu et al [18] proposed a V‐MSER algorithm to extract MSERs from multi‐channel for text detection. Yin et al [20] treated each MSER region as a vertex processing in a graph and then converted text detection into the graph partitioning problem.…”

“…Recently, more efficient text detection models in complex environments have been proposed, such as maximally stable extremal region‐based (MSER) methods [18–21], convolutional neural network (CNN) methods [22–29] and so on. Tian et al [27] proposed a connectionist text proposal network to get accurate text line localisation in a natural scene image.…”