method: NLPR-PAL2017-06-29

Authors: Wenhao He, Fei Yin, Cheng-Lin Liu

Description: this method is based on [1]. We also use images of Chinese and Japanese and Korean from the MLT competition.
[1] Deep Direct Regression for Multi-Oriented Scene Text Detection. arXiv:1703.08289

method: OpenCV + Tesseract2017-09-01

Authors: RRC DOST organizers

Description: Code here

Only modification is the minArea and maxArea parameters of the ERFilter detector (in line 92) in order to detect very small text as follows:

Ptr er_filter1 = createERFilterNM1(loadClassifierNM1("trained_classifierNM1.xml"),8,0.000015f,0.00013f,0.2f,true,0.1f);

Ranking Table

Description Paper Source Code
2017-09-01OpenCV + Tesseract1.55%7.95%2.60%

Ranking Graphic