Parallelization of Levelset-based Text Baseline Detection in Document Images

In this paper, we propose a text baseline detection method. The proposed method is based on a strategy of object separation in a binary image that consists of three steps. The first step is making a binary image with sobel edge detection and mathematical morphology operation to take a approximated text area from the ordinary document image. In the second step, line segments which are candidates for text baselines, are extracted by parallel levelset method. The last step fits a line from each segment with parallel random sample consensus and selects appropriate lines automatically. For parallel computation, OpenMP that is standard API for shared memory parallel programming in C/C++ is used.