基于多特征融合的高动态舞蹈视频关键帧提取系统

Key Frame Extraction System of High Dynamic Dance Video Based on Multi Feature Fusion

由于单一的图像特征并不能很准确的描述视频内容,会导致后续提取出来的关键帧出现漏检和错检问题。针对上述问题,设计一种基于多特征融合的高动态舞蹈视频关键帧提取系统。该系统按照需要给三种不同的特征向量(颜色特征、纹理特征、形状特征)分配合适的权值,然后将特征与权值相乘后再相加,得到一个多特征组合,最后以特征组合作为视觉特征,结合聚类算法,实现高动态舞蹈视频关键帧提取。结果表明:在本文所设计系统的应用下,关键帧提取结果与实际设计的结果相同,没有出现关键帧漏检和错检问题,关键帧提取质量较高。

As a single image feature can not describe the video content accurately, it leads to the problems of missed detection and false detection in the subsequent extracted key frames. To solve the above problems, a high dynamic dance video key frame extraction system based on multi feature fusion is designed. The system allocates appropriate weights to three different feature vectors(color feature, texture feature and shape feature) according to the needs, then multiplies and adds the features and weights to get a multi feature combination. Finally, the feature combination is used as the visual feature and the clustering algorithm is combined to realize the key frame extraction of high dynamic dance video. The results show that: under the application of the system designed in this paper, the result of key frame extraction is the same as that of the actual design, and there is no problem of missing and wrong key frame detection, and the quality of key frame extraction is high.