Rapid-Rich Object Search Lab

Fast Text Image Detection

Devadeep Shyam  , Wang Yan  and Prof. Alex C. Kot



In this project, we propose a text detection method to differentiate images embedded with sufficient text (term as text images) and other images. Since the quantity of text images is much less compared with other images through the internet, this technique could highly reduce governments labor in blocking defamation or illegal commentaries which are conveyed by images. After text image detection, the government agencies only need to focus on a small portion of images. Our technique is able to achieve 99.63% of accuracy in text image detection with a time cost of 0.01 - 0.04 seconds per image under demo software.
Example of Text Images 






Our focus is text image detection, which is the first and key step for blocking transmission of illegal information via text images. Therefore, whether the content of the text is sensitive or not is beyond our concern. While many algorithms have been proposed on text image detection, our main aim is to achieve an automated and scalable solution for the detection and our purpose of this project is to facilitate internet service providers and government agencies to block slander.





Demo of the project and sample images can be download here.