The Method For Extracting Side Page Of 3D Book Model

At present,the distortion of document image optical character recognition(OCR)is still a difficult problem.2D image text recognition rate has a great fall because of the distorted document.The 3D scanning technology for document image distortion appear.It provides a new solution to character recognition problem.However,under normal circumstances,only the 3D model from scanner which consists of two pages can be obtain.The model also includes background space and book fore-edge.And it is a big trouble to segment point cloud artificially.Based on the method for extracting side page of 3D book model,two efficient automatic extraction methods based on morphological characteristics and based on texture are presented.The method based on morphological characteristics use the point cloud depth information.Firstly the background plane equation is being worked out,then the model background is deleted and the books area is kept.Next,by analyzing the shape features of the model books,the space plane to segment two side pages is determined to keep one side page.Finally,the point cloud of book fore-edge is removed based on the point cloud density distribution.Thus eventually the three-dimensional model of the single-side page is kept.At last,the space vertex coordinates index values corresponding to the planar texture coordinate index planar texture coordinate index is modified and the new 3D model file is saved.The method based on texture make the messy texture to orderly texture images firstly.Then it transform the RGB values of pixels to HSI space and a threshold value range is set on the H(hue)and I(luminance),the background model is removed and the books area is reserved;Then through analyzing the spine line in orderly texture image pixel value distribution features,the divided position of the page is found.Based on affine transformation,its 3D coordinates is calculated and the page segmentation space plane can be found out to keep one page.Finally,according to the algorithm described above,the whole design of system is carried and the system is tested for a lot of book models.Experiments show that the two algorithms can steady extract the book pages which tilt to the left,has no tilt,or tilt to the right.And the processing speed and extractive precision can be improved.