中文摘要：

为实现压缩域视频快速检索，提出基于帧数据量波动特性的检索方法．该方法首先计算压缩域各图像帧的数据量，得出查询片段和目标视频等长内的数据量曲线，然后在，帧对齐的基础上将查询片段在目标视频上进行滑动，滑动窗长为单个图组长度．再在每次滑动后计算查询片段与目标视频数据量曲线波动的差异程度，同时每次滑动后要更新目标视频的数据量曲线．最后结合设定门限进行相似判决并返回结果．该方法不需要为每一帧抽取高维特征向量，用一个向量而不是一组高维向量来表述一段视频．实验结果表明，相比现有快速检索算法，该方法使检索速度得到提高，同时也能达到较高的准确率．另外，该方法既可用于基于压缩域视频库的快速检索，也可用于在线的视频片段匹配，实时发现与设定目标相似的视频．

英文摘要：

To implement the fast retrieval for compressed video data, a retrieval method based on the fluctuation of frame data counts is proposed. The data counts for each frame in compress domain are calculated to acquire the data counts curves of equal-length query segment and target video. Then, the query segment is slided on the target video after the alignment of frame data I , and the length of sliding window is the same as the length of a group of picture. The differences of data counts fluctuation between the query clip and the target video are measured. Finally, the similarity result is given according to the designated threshold. The extraction of high dimension feature vector for each frame is omitted in the proposed method, and a video clip is represented by a single vector instead of an array of high dimension vectors. The experimental results show that the video retrieval is speeded based on the proposed method. Meanwhile, a high accuracy is achieved. Therefore, the method can be used for fast retrieval on compressed video database and online video segment matching to find the target video.