In this paper, we propose a method that is able to handle diverse scene images with different font sizes, orientation and background complexity. For this, we use canny edge detection to localize high density location in image and dilate high density pixels by extending pixels environments to create connected components. Then try to blocking each connected component by skeleton structure. The next step result a proper foreground which is separated from the background. Then, in order to output only the text part, some significant features such as projection, corner density and variance are applied to each candidate blocks to recognize them as text or non-text blocks. The experimental results show that our proposed approach is robust enough to face the complex background.