摘要： 为提高视频内容检索方法的鲁棒性，提出一种基于AdaBoost的多特征融合指纹检索方法。通过对样本数据的训练，自适应地获得尺度不变特征变换特征、运动特征以及音频特征的权重，利用得到的权重融合音视频特征，以产生视频指纹。实验结果表明，该方法的准确性较高，在尺度变化、亮度变化、音频噪音攻击下具有较好的鲁棒性。

关键词: AdaBoost算法, 视频指纹, 运动特征, 音频特征, 动态贝叶斯网络

Abstract: This paper proposes a fingerprint retrieval method of multi-feature fusion based on AdaBoost to improve the robust of video fingerprint. The proposed method can gain the weight of Scale Invariant Feature Transform(SIFT), temporal and audio feature adaptively by training the sample data, then fuse audio-video feature to produce video fingerprint according to the weights of the three features. Experimental results show that this method can gain higher accuracy, and have good robustness under various geometric, brightness modification and audio noise.

Key words: AdaBoost algorithm, video fingerprint, motion feature, audio feature, Dynamic Bayesian Network(DBN)

中图分类号: 

• TP301.6