Video Caption Localization And Recognition

With the rapid development of Internet and multimedia technologies,amount of videos rise up explosively on Internet.The network has been heating up and the digital process has been faster and faster,which makes rich online video content including a growing number of important information.Online video services are faster in speed and even more convenient,but also add new security risks.Thus,retrieval and security monitoring for video contents becomes more necessary.Searching manually by keywords or titles is not only inefficient but also unrepresentative.In order to find and monitor the required video information from the vast amounts of data quickly and accurately,it is necessary to understand the contents of the video.Video captions which contain a wealth of high level information,with a strong correlation with the video,are important clues to understand the video contents.Because of low resolution and complex background of online video,video caption cannot be well recognized by OCR.Localization and recognition of Video caption in complex background can help better retrieval of video captions and security of video monitoring,and can also be more efficient translation of video captions,which can greatly improve efficiency and save manpower.Video caption localization and recognition system which consists of two subsystems:video caption key frames extraction and video caption extraction is designed base on this requirement.Firstly,through shot segmentation by histogram difference between frames and considering the various statistical characteristics,the number of video frames which needs caption localization can be decreased by using video shot detection.Then,video caption can be coarsely localized by Sobel edge detection and can be finely localized through morphology and connected components analysis.Finally,binary video caption image which is extracted through Niblack is then recognized by OCR.The system can extract video caption key frames effectively that are representative and localize video caption precisely,and then use OCR for recognition.The experimental results demonstrate that this system can achieve the promising results in video retrieving and editing.