针对如何在镜头基础上进行聚类，以得到更高层次的场景问题，提出了一个基于语义的场景分割算法。该算法首先将视频分割为镜头，并提取镜头的关键帧。然后计算关键帧的颜色直方图和MPEG-7边缘直方图，以形成关键帧的特征；接着利用镜头关键帧的颜色和纹理特征对支持向量机（SVM）进行训练来构造7个基于SVM对应不同语义概念的分类器，并利用它们对要进行场景分割的视频镜头关键帧进行分类，以得到关键帧的语义。并根据关键帧包含的语义概念形成了其语义概念矢量，最后根据语义概念矢量通过对镜头关键帧进行聚类来得到场景。另外．为提取场景关键帧，还构建了镜头选择函数，并根据该函数值的大小来选择场景的关键帧。实验结果表明，该场景分割算法与Hanjalic的方法相比，查准率和查全率分别提高了34．7％和9．1％。

The scene segmentation is a high-level temporal video segment.This paper presents a method of scene segmentation based on semantics.At first,the video clips are segmented into shots and the shot key frames are extracted.Then the features of color histogram and MPEG-7 edge histogram of each key frame are computed and the feature vectors of shot key frames are formed.The support vector machines(SVM) are trained by these feature vectors and 7 binary classifiers in accordance with difference semantic concepts are constructed.These binary classifiers are used to classify the shot key frames of the video clips based on the features of the color and the texture and the semantics concepts of shot key frames can be obtained.The semantic concept vectors of shot key frames are formed by the semantic concepts contained in the key frames.The shot key frames are clustered by the semantic concept vectors and the video scene can be constructed.The shot select function is defined to extract the scene key frame based on the value of function.The experimental results shown that the recall and the precision of this algorithm are higher than those of the Hanjalic's method about 34.7% and 9.1%,respectively.