Finding Text in Natural Scenes by Visual Attention

Text detection is a useful tool for many applications due to the widespread availability of mobile phones and cameras. However, this is considered a complex task for computer vision applications because of the amount of information to be processed. In this work, we present a bottom-up visual attention method appropriate to perform scenery character detection, considering that text regions often present high contrast. In our algorithm, local contrast is evaluated in different scales, using luminance and opponent colors channels as features. Our algorithm outperforms four other state-of-the-art salient regions detection methods in a variety of font types and background conditions.