Document Image Correction Technology Research Based On Registration

The Registration algorithm in image processing has been widely used in practical engineering. Engineering needs are different, so there are also many differences of the image registration algorithms. Fundamental image registration algorithm is divided into three categories:Method based on image gray; Method based on image feature; Method based on model transformation. Registration method based on the image gray by calculating the relevance of the two pictures to choose the result point pixel. Such method is a window template matching method. The method based on image feature need use some of the feature extraction algorithm which first extracts the important information in the image features, and then focuses on these features. Such an approach has been widely applied, it has high efficiency and high noise immunity. Method based on model transformation is a registration through mathematical model, and it is most used in medical field.In the field of distorted document registration, there are a few existing algorithms. For its own reasons, distort document image can not easily be OCR recognized. The distortion causes low OCR recognition. In this paper, we studied the image characteristics of the distorted image, and the text line features of it. We analyzed registration of the distorted documents under different camera angles, and then proposed an automatic registration method of the distorted text documents. First, we need extract text lines from the input text image, then remove the errors points by calculating the sample variance of collection of sample points and use the least squares to fit a parabola. We used the parabola to calculate the inflection point of the image, and limited the registration area. We searched the registration point of the inflection point in another image in the inflection area, and then did the affine transformation synthesis to mosaic the two images. Finally a good result was returned through OCR recognition feedback.The experiment data shows that automatic registration method can return a good image which is mosaiced by the two distorted documents and the image recognition rate has been greatly improved. This method is suitable for libraries to save a lot of distorted electronic files of books.