Text detection from scene images is a challenging topics because of low resolution, complex background and font/font size variations. In this paper, we design a method to detect text based on Naïve Bayes classifier and connected component analysis. We used Naïve Bayes classifier to convert original gray level image into binary image, then connected component analysis is used to identify candidate text regions. In the last step we use empirical rules to determine threshold which used to discard non-text regions and keep the text regions. The proposed method compares between three classifiers outcome; the first is based on Otsu method, the second classifier outcome is derived using Naïve Bayes classifier based on mean feature and standard deviation feature, we named this method Bayes_Two_Features or shortly Bayes2. The last classifier outcome is derived using Naïve Bayes classifier based on just the mean feature, we named this method Bayes_Single_Feature or shortly Bayes1. Otsu’s method is used to convert grayscale image to binary image by assuming that image contains two classes; foreground and background.
Experimental results show that Bayes2 classifier outperforms the other two methods, in the case of big letters especially when these letters are in non-horizontal and skewed form.