Research Of Distorted Document Image Restoration Based On3D Model Technology

Optical Character Recognition (OCR) technology can transfer words from document images into pure text documents automatically after recognition. It could benefit our lives tremendously. Thick books always cause distortions that could reduce the accuracy rate of OCR. To tackle this problem, this thesis introduces a distorted document image restoration method based on reverse engineering. This method restores the distortion caused by the thickness of the book by making the book flat in3D architecture. Experiments showed that the method could effectively flatten the pages in the book and improve the accuracy rate of OCR.At first, the current research results of distorted document image restoration are introduced based on technical classification. Their advantages and disadvantages are given. Then the thesis comes up with the new method of distorted document image restoration based on3D model technology.The first step of this process is to collect the3D model and apply texture on it. With the help of3D scanner, the graphic information is collected and texture information of a book. Then the texture is added on the surface of3D model.The second and third step of the process is projection and stretching. Projection is used to correct the deviation compared to the world coordinate system. Stretching helps to stretch the compressed words in the spine, as a result of which, the outcome of a flatten book is made. Four methods are introduced in order to stretch the book, they are methods of cubic spline interpolation, quadratic fit, cubic fit and discrete algorithm.At last, the thesis introduces a system which was made for this research, which proves the efficiency and robustness of this method.