Text detection via edgeless Stroke Width Transform
Anhar Risnumawan, Chee Seng Chan · 2014
Text detection in scene images has gained widespread interests. A notable work, which is the Stroke Width Transform (SWT), has been attracting much interests due to its simplicity and efficiency. However, the SWT has difficulty in situations such as blur, low contrast, and illumination change images since it highly relies on the outcome from the edge detector. In this paper, a novel method is proposed to obtain stroke width image without the edge detectors. In particular, we replace the edge detector algorithm with the Extremal Regions (ERs) and propose a novel weighted Markov Random Field (MRF) method with three properties to construct a finer stroke width image. Experiment results on ICDAR datasets and a comparison with the state-of-the-art methods have shown the efficiency of the proposed method.