Automatic Reconstruction of Cross-Cut Chinese Document Shreds Based on the Feature of Typesetting and Strokes
Yunqiong Wang, Binggang Wu, Lijin Gao, Hongrong Yang · 2019
This paper proposed an algorithm for automatic reconstruction of cross-cut Chinese document shreds based on the similarities of stroke and typesetting features. We started with extracting the typesetting feature including row height, line spacing and word width based on horizontal projection vector of strokes among the shreds, and clustered shreds into groups by means of k nearest neighbor mutual voting clustering algorithm with rejection region on the basis of horizontal projection vector similarity, word height and line spacing match measurement. Then, vertical sorting among the cluster groups was carried out according to word height and line spacing match measurement. Finally, intra-group splicing was accomplished based on greedy algorithm by using similarity of strokes and word width match measurement on both sides of vertical cut. The experiment result showed that the proposed algorithm achieved 100% reconstruction accuracy with the run time less than 5 seconds, and worked well in environment with 10 standard deviation Gauss noise.