Scene Text Segmentation Based on Local Image Phase Information and MSER Method

The objective of text segmentation algorithms is a pixel-level separation of characters from the image background. This task is difficult due to several factors such as environmental aspects, image acquisition problems, and complex textual content. Up to now, the MSER technique has been widely used to solve the problem due to its invariance to geometric distortions, robustness to noise and illumination variations. However, when pixels intensities are too low, the MSER method often fails. In this paper, a new text segmentation method based on local phase information is proposed. Phase-based stable regions are obtained while the phase congruency values are used to select candidate regions. The computer simulation results show the robustness of the proposed method to different image degradations. Moreover, the method outperforms the MSER technique in most of the cases.