摘要： 提出了一种在视频帧中提取文字区域的算法，该算法基于文字的纹理特征—— 角点，即文字，特别是汉字，含有丰富的角点。首先，在视频帧中进行角点检测，得到视频帧的角点分布图。第2步是进行角点滤波，即在角点分布图中剔除一些比较分散的，可以认为不是从文字中检测到的角点。第3步对角点分布图中余下的角点聚类形成候选文字区域，采用水平方向和竖直方向的投影对这些区域进行分化。第4步文字区域的最后确认，除去噪声。实验表明该算法计算速度快，提取文字区域准确。

关键词: 文字区域提取, 角点检测, 投影

Abstract: An algorithm for finding text in video frames is proposed based on the texture character of text —— corners that the text,especially Chinese text,are rich of corners. First,corners detection is used in video frames,getting the corner map. Second,corner refinement,eliminating the isolate corners and the corners not detected from text. Third,the remained corners in the corner map are merged to form the candidate text region,horizontally and vertically projecting are used to decompose the candidate region. Finally,text region verification is taken to reduce false alarms. Experimental results show that the algorithm is accurate and spend little time.

中图分类号: 

• TP391.4