원문정보
초록
영어
Comparing to other videos, text in news video contains more semantic information which are mostly descript the news event. So, the extraction of text plays an important role for obtaining the high-level semantic information from the video. In this paper, the improved Trajkovic corner detection algorithm has been studied based on the characteristics of text in news video. According to the characteristics of text, the adaptive thresholding method of scale in combination with standard deviation (SCSD) can be applied to determine the corners, which aims to accurately obtain the text-related corners. The breadth-first clustering algorithm was also been used to distinguish and plan the detected corners which are in same range to determine the text area in video frame. The accuracy of recognition and semantic exploration of the extracted text can be used to quickly understand news video that can effectively improve the efficiency to summarize the news videos. The experimental results show this method with high pertinence which can better achieve the targets. The detection accuracy of text-related corners can reach to 82.1% which is higher than other methods. Under the same experimental condition, recall ratio and precision ratio of the text area are also improved dramatically.
목차
1. Introduction
2. Previous Researches
2.1. Method Based on Region
2.2. Method Based on Texture
2.3. Method Based on Region and Texture
3. Positioning And Extraction of Text Area
3.1. Caption Frame Detection
3.2. The Improved 8-neighbors Trajkovic Corner Detector
3.3. Local Non-maximum Suppression of Corners and Text Area Positioning
4. Experimental Analysis
4.1. Corner Detection Results
4.2. Text Area Positioning Results
5. Conclusion
References