Abstract: A novel content-based clustering algorithm is presented for video emotion types recognition in this paper.The scene brightness，scene rhythm，movement intensity and color energy in a video scene are selected as the low-level video features.The methods of data extraction from each emotion feature are presented in detail，and the video emotion feature vector is created accordingly.After discussing the video object space and the emotion measure function，the unascertained clustering model of video scene is built.A method is applied to specify index weight of each emotion feature and perform video samples clustering.The experimental results verify the feasibility and effectiveness of this algorithm.

Key words: video retrieval, affective content, unascertained cluster, features weight

摘要： 提出了一种新的视频语义分析算法，着重对情感类型识别进行了聚类研究。选取场景亮度、场景节奏、运动强度和颜色能量作为视频情感低层特征，详细介绍了每种情感特征的数据提取方法，并构建了视频情感特征向量。在分析了视频对象空间和情感测度函数之后，建立了未确知视频场景聚类模型，给出了情感向量各分量权重确定与视频样本聚类的方法。仿真实验数据验证了未确知聚类方法对视频情感内容的识别是有效的、可行的。

关键词: 视频检索, 情感内容, 未确知聚类, 特征权重