Labeling-and-DUDE-based Denoising Method for Binary Image

In this paper, we propose a method to accurately denoise binary images by capturing wider range of neighborhood information with reducing the computational complexity. The proposed method first removes isolated pixels from the entire image by preprocessing with labeling. Then, the DUDE algorithm, which is commonly used for denoising binary images, is applied multiple times with different input information. The results are used to denoise the noise that occurs at the boundary between the text and background regions. Additionally, we propose parameter setting method which is suitable for the proposed method. In the experiment, we use image data of document images with added noise. Comparison of the proposed method with conventional binary image processing methods shows the effectiveness of the proposed method.