Road sign text detection using contrast intensify maximally stable extremal regions

This work focuses on the text detection of road sign directional board from the outdoor environment. We propose a fast and effective method to detect texts in the natural image and remove the blurring problem by adding a contrast enhancement method with Maximally Stable Extremal Regions (MSERs). Character candidates are detected by a MSERs algorithm with contrast intensify method. After that, non-text regions are removed with the geometric rules such as aspect ratio. Then to remove the false positive, the stroke width variation approach imposes. The properties of character candidates (e.g. stroke width, intensity, size, etc.) are used to form the word and distance between the characters are measured by the Euclidean Distance algorithm. Finally, text candidates are identified by the Optical Character Recognition (OCR) and send a message to the drivers or pedestrians. This method has been evaluated by the public data set ICDAR 2011, ICDAR 2013, ICDAR 2015 and also a set of road sign directional board images.