Apply SOM to Video Artificial Text Area Detection
Apply SOM to Video Artificial Text Area Detection
复制标题
将SOM应用于视频人工文本区域检测
DOI：
10.1109/icicse.2009.13
复制
发表时间：
2009
期刊：
影响因子：
--
通讯作者：
Yan Wang
中科院分区：
文献类型：
--
作者：
Jia Yu;Yan Wang
视频人工文本检测是模式识别中的一个具有挑战性的问题。当前通常基于边缘、纹理、连通域、特征或学习的方法总是受到视频中人工文本的大小、位置、语言的限制。针对上述问题，本文将基于监督学习的SOM（Self-Organizing Map）应用于视频人工文本检测。首先，提取文本特征。并考虑到视频人工文本的局限性，采用人工文本每个像素的位置和梯度作为分类特征。然后提出了三层监督SOM来对视频图像中的文本和非文本区域进行分类。最后利用形态学运算得到了更加准确的文本区域结果。实验表明，该方法可以有效地定位和检测视频中的人工文本区域。
Video artificial text detection is a challenging problem of pattern recognition. Current methods which are usually based on edge, texture, connected domain, feature or learning are always limited by size, location, language of artificial text in video. To solve the problems mentioned above, this paper applied SOM (Self-Organizing Map) based on supervised learning to video artificial text detection. First, text features were extracted. And considering the video artificial text's limitations mentioned, artificial text’s location and gradient of each pixel were used as the features which were used to classify. Then three layers supervised SOM was proposed to classify the text and non-text areas in video image. At last, the morphologic operating was used to get a much more accurate result of text area. Experiments showed that this method could locate and detect artificial text area in video efficiently.