Text Region Extraction in High Contrasting Image
Seokjun Kang, Daewoong Cha, Youngwoo Kim, Dong Seog Han · International Journal of Future Computer and Communication · 2017
In this paper, a text region extraction system with high contrasting text images for self-driving cars is proposed.The maximally stable extremal regions (MSER) method is usually used to extract text regions.Images must be converted to grayscale to process with the MSER method.However, the performance of MSER by using grayscale images has a poor ability of capturing regions of interest in bad conditions such as high-contrast, low-luminance, much light reflection, and so on.An MSER system with a contrast-limited adaptive histogram equalization (CLAHE) instead of conventional MSER is therefore proposed.CLAHE is utilized as a pre-processing method in MSER to detect text regions.The proposed method achieves a precision of 81% and a recall of 82%.However, those for the MSER with grayscale are 63% and 55%, respectively.