Character Image Preprocessing and Matching for Writer Identification
Ai La · Zhongwen xinxi xuebao · 1996
In writer identification(WI) ,writing styles are always compared by character match-ing ,while image preprocessing and normalization are very important for matching. In this pa-per, the techniques of preprocessing and shape matching for writer identification are presented. For preprocessing the noise elimination and size normalization are treated in this paper. The noise of character images is eliminated via smoothing,contour tracing and filling. The size nor-malization is linear in order to preserve the writing features. And the character position and scale are determined by three approaches ? square bounding, centroid alignment and single bounding. After normalization, writer recognition is implemented by character shape matching using distance transform(DT)which is computationally simple and distortion tolerant. The ex-perimental results show that centroid aligned normalization is the most suitable for writer iden-tification. The recognition rate is satisfactory which demonstrate the effectiveness of DT matching.