Caption detection and text recognition in news video
Zhe Yang, Ping An Shi · 2012
A new method for caption detection and text recognition in news video is proposed. The color feature is used to detect the candidate caption frames and the edge feature is used to filter out the wrong frames. Then, the pre-processing which includes interpolation and binarization is applied to the caption images before they are sent to the Optical Character Recognition (OCR) engine. The experiment results show that the proposed method has a better performance in accuracy of caption detection and text recognition.