Robust Binarization of Non-Uniform Illuminated Images

Image binarization is a basic image processing technique with many applications. Binarization methods have been incrementally improved to deal with real world conditions. A particularly important consideration is non-uniform illumination, present when using general purpose devices for image acquisition (e. g. mobile phones). In this paper we present a new binarization method for non uniformly illuminated text images. The main idea is to improve illumination using two complementary illuminated images. To achieve this we estimate the luminance of the original image and we create a new complementarily illuminated image. We then fuse the two images and for each pixel we estimate in which of the two images the pixel is correctly illuminated. The proposed method is evaluated with synthetic and real images, using several metrics, showing better performance than state-of-the-art techniques.