An Improved Drop-fall Algorithm Based on Background Analysis for Handwritten Digits Segmentation

Among several contour splitting algorithms, Drop-fall algorithm has shown good performance, but it is defective in locating the start point of the algorithm and the seeping process in vertical direction inside character stroke block. For the purpose of solving the problems produced by the defections, in this paper Drop-fall algorithm is improved by citing water reservoir concept to use the information coming from the background region of touching digits. On the basis of the topological features of background region, the range of searching start point in algorithm is dictated and the optimal position for the seeping process is determined. For connected digits sharing a long stroke segment, instead of the seeping process in vertical direction, centerline is extracted from common touching area to be used as a long tilt path which is cutting across the black block when the drop enters the junction of connected digits. Experimental results show that the improved algorithm obtains more satisfied segmentation and works better than former.