基于哈希学习的舞蹈视频中特定动作编码与检索

Specific action coding and retrieval in dance video based on Hash learning

针对传统动作检索方法在处理数据冗杂度高、特征维度多的舞蹈视频图像时,出现特定动作的编码和检索耗费时间长、效率低等问题。文中提出了一种基于哈希学习的舞蹈视频中特定动作编码与检索的高效方法。该方法首先通过哈希函数将原始高维图像数据映射到低维空间,然后利用迭代量化将得到的低维向量量化成该帧图像所对应的哈希编码,最后通过计算与动作编码库中的标准哈希码间的海明距离确定其对应的特定动作,进而完成对特定动作的编码与检索。试验结果表明,文中所提出的编码与检索方法相较于传统方法在处理速度与检索准确度上均有显著提升,从而验证了该方法的可行性。

Aiming at the problems of high data complexity and multi feature dimension in traditional motion retrieval methods,such as time consuming and inefficient encoding and retrieval of specific actions,this paper proposes an efficient method for encoding and retrieval of specific actions in dance video based on hash learning.Firstly,the original high dimensional image data is mapped to low dimensional space by hash function,and then the low dimensional vector is quantized by iteration quantization to the corresponding hash coding of the frame image.Finally,the corresponding specific actions are determined by calculating the Hamming distance between the standard hash codes in the action coding library.Coding and retrieval of specific actions in pairs.The experimental results show that the coding and retrieval methods proposed in this paper have greatly improved the processing speed and retrieval accuracy compared with the traditional methods,which verifies the feasibility of this method.