High-accuracy rectification technique of deformed document image using Tiled Rectangle Fragments (TRFs)
Summary
Digital books are getting popular because of the spread of electric devices such as ebook reader, smartphone and tablet. Book digitizaiton technologies play significant roles in supporting such popularity. However, traditional way of digitization that digitize books with flatbed scanners is time-consuming and needs human power. To overcome this obstale, we have developed a novel system called Book Flipping Scanning (BFS), which digitizes whole a book within a minute. The key point in achieving high-speed digitization is to perform scanning while the book is flipping. While it enables to scan very quickly, this stype of book digization, scanning during flipping, naturally raises rectification problem because the captured image is deformed.
This study approaches such a rectification problem by introducing a novel model 'Tiled Rectangle Fragments' (TRFs) representing a developable surface (details follow.)
Rectification problem is to construct a 2D to 2D mapping from a warped document image to its rectified document image. Revisiting this problem from 3D point of view, it can be understood to construct an isometric mapping from 3D deformed document surface to its developed surface. TRFs represent a developable surface by tiling the same size of rectangle. Each rectangle is a subdivied fragment of the developed surface of the original developable surface. This model enables us to construct a quasi-isometric mapping and rectify with high accuracy.
Note that the proposed method only requires 3D point cloud of the deformed document surface and its captured image. This means that our method doesn't depend on the texture on the document.
Reference
- Masahiro Hirano, Yoshihiro Watanabe and Masatoshi Ishikawa: 3D Rectification of Distorted Document Image based on Tiled Rectangle Fragments, 2014 IEEE International Conference on Image Processing (ICIP2014), (Paris, France, 2014.10.29)/Proceedings, pp.2604-2608 (2014)
