Distorted Document Image Mosaicing Technology Research Based On Feature-point Matching

Distorted document image correction technology is a hot spot of document image deformation correction technology, its role is that the OCR recognition rate of distorted document images can be obviously raised after processing of distortion correction technology. Currently, distorted document image correction technology mainly contains the method based on model, the method based on3D and the method based on document image content segmentation. Those methods focus on correcting one distorted document image and achieve the effect in a certain degree. However, if there are seriously distorted text areas in the document image, the methods will be hard to correct the image because of the serious deficiency of the text information.Aiming to the circumstance, this paper adopts image mosaicing technology to correct the distorted document images, avoids using the distorted area of the document image but directly mosaics the undistorted and content-complementary text areas of the two images, so then produces a completely new and content-complete undistorted document image. Besides, this paper adopts a method of inflection position automatically detection in order to implement the document image automatic mosaicing.Specifically, the paper first takes pictures of one page from left and right angles, produces two distorted document images, executes pre-processing operation on these two images including gray processing, binarization and morphology dilation, chooses a image as the base image and then extracts the middle lines of the text lines, acquires the distortion position of the document image according to the descending extent of the first middle line, sets two inflection feature points as the base point pair from this inflection position, then makes the registration processing between two gray images combining the methods of area limitation, vector constraint and similarity measure based on gray value. At last, mosaicing the two images according to the affine transformation model.As the result of the experiments show, after using the automatic inflection detection and image mosaicing technology to make the correction, the OCR recognition rate has been raised obviously, meanwhile, the methods of area limitation and vector constraint play a certain role in increasing the speed and accuracy rate of the image registration, which enhances the practicality and robustness of the automatic image mosaicing technology.