Writer Identification by Multichannel Decomposition and Matching
基于多通道分解与匹配的笔迹鉴别研究

By writer identification (WI), the writer of a handwritten document is detected by analyzing the writing style. The crucial stage in WI is the extraction of individual features which reflect the writing style. There are many informative individual features, in which are the stroke position, direction, and the collocation. Some features can be represented simutaneously by multichannel decomposition (MCD) of images. In this paper, a MCD approach of binary images for WI is presented. The image is decomposed into directional subimages according to strokes direction, and then each subimage is further decomposed into several frequency bands. Finally, the sampled values of subband images are used as individual features for WI. Promising results have been achieved in the writer recognition experiments by feature matching.