Абстрактный

Local fragment distribution features for text-independent writer identification

Ding Hong, Yang Feng-ying, Zhang Xiao-feng

In this paper, an efficientmethod for text-independentwriter identification using Local Fragment Distribution Feature (LFDF) is proposed. Local fragments, which are parts of the contour in sliding windows, contain the information of strokes. Our method uses the distributions of to create LFDF vector for each specific manuscript. In order to reduce the impact of stroke weight, the fragments which do not directly connect the center point of the sliding window are ignored. Then, the distributions of local fragments are counted and normalized into LFDF. At last, weighted Manhattan distance is used as similarity measurement. The proposed method offers state-of-art performance on ICDAR2011writer identification database with multi-languages and the experiments demonstrated that this method is suitable for text-independent writer identification.