In this paper, we focus on the binarization methods as a core step in most image processing
algorithms especially localization of the characters in scene images. We have developed in this
paper our previous scheme which based on shape properties and geometric features to define text
region and adopt our binarization scheme which based on Naïve Bayes classifier to convert
grayscale image to binary image. Then we compare this binarization scheme with four famous
different methods and explore their effects on detection characters in scene images. We found
that our method outperforms the other four prior methods in detection characters with respect to
Recall metric and the Otsu method follow our methods.