Simulating inertial and centripetal forces for segmentation of overlapped handwritten digits

This paper presents a new algorithm for segmentation of overlapped cursive handwritten digits. Segmentation is a common issue for automatic character recognition. The work in digits is common in applications as bank check or postal code processing. Our method is based on the segmentation achieved by the path traversed by a hypothetical ball which rolls over the digits under the influence of inertial and centripetal forces. The proposal has a computational cost of O(n²) instead of O(2ⁿ) as in classical graph-based approaches.