A Video Text Location Method Based on Stroke Extraction and Conditional Morphology

Texts in video frames are powerful sources of high-level semantics. They can be used for video analysis and content-based retrieving. Text location, which is the first and the most important step of text information extraction, affects the following recognition results considerably. In this paper, we strive to propose a text location method based on stroke extraction and conditional morphology. The stroke map of the input image is first got by a stroke extraction operator. Then, to remove the non-text disturbances in the stroke map, we introduce an improved method of morphology: conditional morphology. Compared with the original method, conditional morphology can not only remove the non-text noises but also enhance the text information, which improves the location performance remarkably. At last, the precise location of texts can be obtained by some merge-split rules with a combination of connected component analysis method. Experimental results show that our approach performs well with high speed and precision.