基于角点检测和自适应阈值的新闻字幕检测

目前用于提取新闻视频帧中字幕的方法准确率和检测速度普遍较低，尤其对于分辨率和对比度较小的标题文字，检测效果很差。针对上述问题，提出一种基于角点检测和自适应阈值的字幕检测方法。该方法利用角点检测确定标题帧中的文字区域并进行灰度变换，利用自适应阈值的方法对其进行二值化，得到OCR可识别的文字图片。实验表明，该方法可以快速有效地提取出分辨率和对比度较小的新闻视频标题字幕。

News Captions Detection Based on Corner Detection and Adaptive Threshold

The methods of extracting captions in news videos are always with low accuracy and efficiency, especially for caption text with low resolution and contrast. This paper proposes a detection method based on corner detection and adaptive threshold to solve the problems. It locates the text area based on corner detection, magnifies the image of text area, calculates the gray image and binaries it by using adaptive threshold and feedback method. Optical Character Recognition(OCR) are used to recognition the text in the binary images. Experimental results show that it is effective to extract the text with low resolution and contrast.