Double-edge-model based character stroke extraction from complex backgrounds

Global gray-level thresholding techniques such as Otsupsilas method, and local gray-level thresholding techniques such as adaptive thresholding method are powerful in extracting character objects from simple or slowly varying backgrounds. However, they are found to be insufficient when the backgrounds include sharply varying contours or fonts in different sizes. In this paper, we propose a double-edge model insensitive to stroke width to extract character strokes with an unknown stroke width from complex or sharply varying backgrounds. Also, we propose a novel postprocessing method combining 2-level global thresholding and Canny edge detection to keep the character object in integrality and remove the background simultaneously. Experiment results show that the proposed method can extract character objects from complex backgrounds with satisfactory quality.