| With the explosive growth of news video data, it has become increasingly important to classify, retrieval and manage the massive news videos. The text caption is part of important information contained in news video, and it can provide a wealth of high-level semantic information and help users to understand the video content more easily. Therefore, it is an effective method to analyze and understand the video content by accurately recognizing the video caption.Considering the features of complex and constantly changing background as well as strong global noise of the text caption in news videos, in order to accurately locate text caption region in news video, we adopt a caption location method based on edge detection and projection. For the problem of bad binary result and low recognition rate by handling a picture of a whole caption region, we adopt a character segmentation method based on gradient projection. Aiming at the problem of unsatisfactory caption recognition result owing to there is no specific OCR language library of news video caption at present, we use the open source OCR engine Tesseract to train the sample set of character images to get the OCR language library used for news video caption recognition. Experimental results show that the adopted methods can effectively locate news video caption region and accurately segment characters in it, and we find that it can effectively improve the recognition rate of news video captions by using the OCR library, so our research has a good practical value. |
PDF Full Text Request