| Binarizing document images with poor contrast, strong noise, complex patterns, and variable modalities in the gray-scale histograms is a challenging problem. A new binarization algorithm was developed to address this problem. The proposed algorithm seeks a threshold that would eliminate background noise, while preserving as much handwritten stroke data as possible. The main advantage of this approach is that it optimizes the binarization of a part of document image that suffers from noise interference, referred to as Target Sub-Image (TSI), by using information easily extracted from another noise-free part of the same image, referred to as Model Sub-Image (MSI). Simple spatial features extracted from MSI are used as a model for handwritten strokes. This model captures the underlying characteristics of the strokes, and is invariant to the handwriting style or content. Since this model is image dependent, it will compensate for variations resulting from using different types of pens, and variations caused by changing image-capturing settings, such as brightness or contrast settings. This model is then utilized to guide the binarization in TSI. The algorithm was developed within cheque processing application environment. We then generalized this approach so that it can be applied to other types of document images. The proposed algorithm provides also a new technique for the structural analysis of document images, which we call “Wavelet Partial Reconstruction” (WPR). Experiments with 9,300 real cheque images showed significant improvement in the binarization quality compared to experiments with other well-established algorithms. |
