Auto determination of distortion parameters for correction of scanned book image

To correct the distortion of scanned book image, a polynomial is proposed to describe the relation between g(zi) and zi, where g(zi) is the grayscale of a pixel and ziis the distance from the pixel′s corresponding point to the scanner′s operation plane. With an image of a twisted plane object whose distortion parameters are known, zican be calculated and the polynomial can be established by the leastsquare method using the actual grayscale gi of every pixel. The experiment indicates, accuracy of the polynomial is good enough when q=4, and every coefficient is computed for later use. For distorted image of plane object twisted at will, distortion parameters are calculated automatically as follows, distortion parameters are estimated by using grayscale of pixels located at the book margin, and the estimation of zi is calculated, and zi is used in the established polynomial to calculate g(zi); the optimal distortion parameters are found by adjusting the parameters until every calculated g(zi) is close to gi. In the experiment, good image correction results are achieved with the parameters obtained by this method. The maximum error is reduced from 41 percent to 6.9 percent. Thus, image correction can be done automatically without manual operation.