Text detection plays an important role in the system of text information extraction. To solve the problem of detecting text in the bad settings such as complicated Text-background

uneven illumination and so on

a generation text detection method based on stroke direction map is proposed

which utilizes a coarse-to-fine framework. In the coarse step

it firstly builds the stroke direction Information map of the source images with Haar wavelet and LBP descriptor. Then the candidate text lines can be obtained through filtering

component analysis and projection profile analysis of the direction Information map. In the final step

it makes use of SVM classifier with kinds of text texture features to verify the real text lines. Experiment results on ICDAR03 show that the proposed algorithm can quickly and accurately detect texts region in different settings.