A Fast Approach for Text Region Detection from Images on Online Social Media

Microblogging sites are very popular sources of real-time information, which includes both textual information, images, and videos. Since individual posts on such sites are very small, they can convey only a small amount of information. Hence, in situations like an ongoing emergency or disaster, users wishing to share a large amount of information often resort to including the text in an image and then sharing the image. Utilizing such textual information within images requires a text-detection mechanism that not only needs to be accurate, but also very fast in order to process the hundreds of images posted on social media in real-time. In this work, we propose such a text-detection algorithm from images. Experiments over images posted on Twitter during a recent disaster event show that the proposed method achieves competitive accuracy with a state-of-the-art method, while being much faster.