Method For Skew Degree Detection Of Camera Captured Document Image

Character Recognition from document image is an important image processing technology. The correct rate of character recognition has a great relationship with the quality of the document image. Last years, it is always getting document image by desk-top scanner. The increasing availability of high performance, low-price, portable digital imaging devices has created a tremendous opportunity for supplementing traditional scanning for document image acquisition. Digital cameras attached to cellular phones, PDAs, or wearable computers, and stand alone image or video devices are highly mobile and easy to use. They can capture images of thick books, historical manuscripts too fragile to touch, and text in scenes, making them much more versatile than desktop scanners.Compare with desk-top scanner, document image shot by camera always suffers distortion. For improving character recognition rate, pre-processing is needed. One of the important parts is skew dewarping. It needs skew detection. In this paper, we propose a new method detect document image skew by detecting the variation of black pixels’ number with a short line which rotates on a moving point.At first, transfer original document image into binary image, then make matrix transform processing, divide the image into6parts. At the center of each part, set a point as initial point across which with a short straight line. Rotate the line on the point to detect the black pixels which pass through it. According the variation of black pixels’number, text line skew should be calculated. With the6skew angles, the horizontal vanishing point will be found.The innovation points of this research lie in1. Improve the method finding skew angle with text flow. This method which scanning black pixels with a single pixel width line can getting the skew angle in document image which has extremely narrow space between text lines.2. Without pre-assuming the shape of the undistorted image of text, it is able to get the correct results.3.No matter it is all or part of text which is shot in image, the method can get the correct result.4.According to the different distance between point and text, moving the point in different distance at once, it saves time of moving points, reduces calculation time, improve the detection efficiency.This method solves following some problems:(1) If the skew angle is large, the accuracy can not be guaranteed;(2) If the image has high complexity space, the calculation is large;(3) The entire document page should be shot in the image;(4) The document image should not has very narrow space between text lines.