Research On Digital Video Semantic Content Extraction

With the rapid development of network technology, the "Internet" not only plays an important role in personal daily life, but also brings the very great effects on official and business activities. However in information era, there exists some bad information, which seriously and harmfully affects the society. Therefore, information content security gets the great place in our social life and national security.Video data contains a plenty of semantic content. Video semantic analysis is an important research work in multimedia information content security. As a high-level semantic clue of video, the content of the caption text in digital videos provides valuable information for people to understand the semantic content of the video. Algorithms for caption text detection, extraction, enhancement and recognition as well are the key techniques for video semantic understanding.Generally the most existing approaches of video caption detection merely focus on the detection of caption text with the aspect of spatial boundaries. This thesis presents an improved video caption detection framework which not only detects location of captions but also determines their temporal boundaries. The framework consists of six main steps: multiple frames integration(MFI), edge detection, iterative text line decomposition, temporal boundary detection of caption, binarization, and OCR as well. Experimental results on several hours video show the efficiency of the framework, which achieves the accuracy of about 80%.