text detection method

texture based method

1.zhong et al. proposed a method for text localization in color images.

2.Li et al. introduced a text detection system to detect and track texts in videos.

3.Kim et al. trained a SVM classifier to classify each pixel by directly using the raw pixel intensity as local feature.

4.Lyu et al. proposed a coarse to fine multiscale search scheme,

5.Zhong et al. proposed an interesting algorithm that can ditectly detect text in the discrete cosine transform domain

6.Chen et al. proposed a fast text detector.The detector is a cascade Adaboost classifier

7.Wang et al. orioised a method for locating specific words from natural scenes.

Component based methods

1.The method proposed by Jain et al. decomposed images into several non overlapping components by color vlustering,and then removed non text components according to geometric rules.

2.Epshtein et al. proposed  a new image operator Stroke Width Transform(SWT)

3.Neumann et al. proposed a text detection algorithm based on MSER, and eliminates invalid candidates using a trained classifier.

4.Zhao et al. constructed a sparse dictionary from training samples and used it to judge whether a particular area in the image contains text.

5.the approach proposed by Yi et al. can detect tilted text in natural images.

6.Shivakumara et al. also proposed a method for multioriented text detection.

7.Based on SWT, Yao et al. proposed an algorithm that can detect texts of arbitrary orientations in natural images.能够识别多角度

8.Huang et al. presented a new operator based on SWT,called Stroke Feature Transform(SFT).在水平方向上有很好的识别效果

9.In,Huang et al. proposed a novel framework for scene text detection,which integrated Maximally Stable Extremal Regions(MSER) and CNN.很好的表现

Hybrid methods

1.In the method propsed by Liu et al. ,edge pixels of all possible text regions were extracted using an elaborate edge detection strategy...

2.the hybrid method proposed by Pan et al. extracts candidate components from multi-scale probability maps.