An Adaptive Method for Thresholding of Document Images

Th resholding is a pre-processing step that is usually used in applications such as automatic text recognition. Some conditions such as environmental lighting can defect segmentation procedure. Adaptive thresholding is usually used to overcome these defections. In this paper, a new method, based on within-class calculation, is proposed to extract text lines in an uneven background. Here it is shown that in many cases within-class variance is a good criterion for finding a proper window size in adaptive thresholding.