The last decade has witnessed a tremendous growth in amount of multimedia data available in the form of images, videos and audios. With this has increased the need to develop efficient indexing and retrieval techniques. Talking about videos, in addition to the video content itself, a powerful attribute for video indexing is the artificial text appearing in it.
The artificial text also known as superimposed or caption text has attractive applications like keyword-based video search and automatic video tagging etc. Over the years, a wide variety of text detection/localization techniques have been proposed achieving promising results. Mostly, these methods have been tested on private data sets and they differ in evaluation metrics as well. As a result, there was a need to develop a standard shared data set with labeled ground truth so that algorithms could be evaluated and compared in a meaningful way.
No comments:
Post a Comment