Learning To Binarize Document Images

Document images produced by cameras often have varying degrees of brightness. To resolve the problem, we propose a method that divides an image into several regions and decides what binarization action to take on each region based on the rules that are derived from a learning process. Since each region can allow more than one action to take, we are dealing with a multi-label and multi-class classification problem that can be solved effectively by support vector machines. Tests on images produced under normal and inadequate illumination conditions show that our method yields better OCR performance than three global binarization methods and four locally adaptive binarization methods.