Text Detection in Images and Videos

The goal of a multimedia text extraction and recognition system is filling the gap between the already existing and mature technology of Optical Character Recognition and the new needs for textual information retrieval created by the spread of digital multimedia. A text extraction system from multimedia usually consists of the following four stages: spatial text detection, temporal text detection – tracking (for videos), image binarization – segmentation, character recognition. In the framework of this PhD thesis we dealt with all the stages of a multimedia text extraction system, focusing though on the designing and development of techniques for the spatial detection of text in images and videos as well as methods for evaluating the corresponding result. Two methods for the evaluation of the text detection result were proposed that deal successfully with the problems of the related literature. Each of them uses different criteria while both of them are based on intuitively correct observations. Finally, a very efficient method was developed for the temporal detection of text which actually conduces to a better spatial detection while concurrently enhances the quality of the text image.