Preprocessing of Handwritten Date Images on Chinese Cheque

The preprocess procedure of handwritten date images in our automatic Chinese cheque processing system is introduced. It consists of three major components: preprinted characters removal, binarization and extra strokes removal. A pixel-contour distance based method is proposed to remove preprinted characters while preserving the connection of object strokes. In binary, a local feature called morphological double edge is used to enhance the character strokes, and then a global threshold is selected by a new recursive method. Extra strokes locate below date characters. We remove them by contour analysis. Experiment results on real-life cheque images demonstrate the efficiency of our proposed methods