基于证据理论的视频语义概念检测

摘要

视频语义概念检测是跨越“语义鸿沟”问题,实现基于语义的视频检索的前提.本文提出了一种基于证据理论的视频语义概念检测方法.首先,分别提取了镜头关键帧的分块颜色矩、小波纹理特征和边缘方向直方图特征；然后,利用支持向量机(Support vector machine,SVM)对3种特征数据分别进行训练,分别建立分类器模型；再次,对各SVM模型泛化误差进行分析,采用折扣系数法对不同SVM模型输出的分类结果进行修正；最后,采用证据融合公式对修正后的输出进行融合,把融合结果作为最终的概念检测结果.实验结果表明,新方法提高了概念检测的准确率,优于传统的线性分类器融合方法.%Video semantic concept detection is a prerequisite to solve the 'emantic gap' problem and realize semantic-based video retrieval. A video semantic concept detecting method based on the evidence theory is proposed. Firstly, features including grid color moment, wavelet texture and edge direction histogram are extracted from the key frames of video shots. Then, for each type of feature, an SVM model is trained. Thirdly, by analyzing the generalization error of each SVM model, a discounting coefficient method is implemented to modify the classification results of these models. Finally, these modified results are fused with an evidence fusion equation, and the fused result is regarded as the final semantic concept detection result. Experimental results show that the new method has improved the detection accuracy and outperforms the traditional linear classifier fusion method.