Degraded document image binarization combining local statistics

In this paper, an efficient binarization algorithm for degraded document images is proposed. The proposed method contains two procedures: background estimation and foreground detection. The background is estimated by median filtering. The rationality is the characters can be assumed as irregular pixels. Text regions are detected by background subtraction and local variance calculation, and then local linear transform is performed to the detected text regions. Finally, a naive thresholding algorithm, ex. Otsu's method, is performed to obtain the binary result. Experimental results show that the proposed method achieves promising results.