Noise removal from binarized text images

Noise is common in binarized images which are the result of extracting text from the embedded text in an image. It degrades the performance of character recognition module. In this paper, a robust algorithm is proposed to eliminate noise from the binarized text image based on the text-stroke width information. First, salt-pepper-like noises are eliminated by a morphological filter, which is to enhance the correctness of estimating the text-stroke width. Finally, a method based on the text-stroke width information is proposed to extract text from noise images. Experiments on a wide variety of binarized images with salt-pepper noise, and cluttered noise reveal the feasibility and effectiveness of our proposed approach in removing the noise.