Skew Angle Estimation and Correction for Noisy Document Images

Document skew commonly occurs during document scanning; it should be avoided because it dramatically reduces the accuracy of the OCR. Noise removal is an important procedure before on going further processing. This paper describes an approach towards noise removal, skew detection and correction for text in scanned documents. Preprocessing is a stage, comprising number of adjustments in order to obtain the noise reduced results, and then the skew angle is estimated. Instead of deriving a skew angle from the text lines, the proposed method uses various types of visual content of image skews, and HDT algorithm is used to select the useful image region dynamically. A bootstrap estimator is finally employed to combine various cues on local image blocks. Once the skew angle is being estimated it has to be rotated in the opposite direction in order to correct the skew angle.