【作者】 夏国武；

【导师】 张曦煌；

【作者基本信息】 江南大学 ， 计算机应用技术， 2008， 硕士

【摘要】 随着数字化图书馆和大型多媒体数据库的广泛使用。视频检索将成为重要的研究领域。最大程度地提高基于内容的多媒体的应用,也正是MPEG-7的目的所在。小波分析是继Fourier分析之后新的时频分析工具,它在科学研究和工程技术中的应用非常广泛。虽然小波理论现已比较成熟,但是近年来有关它的应用研究仍在不断发展更新。小波变换在图像处理领域中的应用几乎可以囊括图像处理的所有方面。本文对小波变换在视频检索方面进行了研究,在分析了这些方面研究现状的基础上,探讨了自己的算法和思想,并给出了相应的实验结果和分析,以下是本文主要研究内容:(1)由于视频序列可以看成帧内二维和时间上的一维,因此可以把已有的图像检索的方法应用到视频检索中。首先利用基于小波的颜色图像检索算法对视频库进行初步筛选,然后根据集合的相似性度量原理对查询片断和目标片断进行相似性的计算,相似度值最大的则表示查询片断和目标片断完全一样,同时该目标片断就是所要求查询的片断。通过实验分析,该方法不仅可以获得较高检索效率,而且有着更高的检索速度。(2)从另一个角度来看,又可以把视频数据看一个三维信号,然后对其进行三维小波变换分解,该方法克服了传统视频检索方法中未考虑的时空关系。首先利用低频子图小波系数的标准方差均值对目标视频库进行初步筛选,然后利用八叉树算法提取细节特征,进一步的把细节特征和近似特征组合起来对图像进行精确的检索。实验证明,该方法具有更好的检索精度,但是由于视频数据量比较大,因此利用此方法在速度上将明显降低。更多还原

【Abstract】 Video retrieval is emerging as an important research area with application to digital libraries and multimedia databases. The objective of the MPEG-7 standard is to enhance content-based multimedia applications to the maximum.For another, wavelet transform is a new tool of time-frequency analysis after Fourier analysis and is widely applied in the scientific research and project technology. Though wavelet theory has been somewhat perfect, its applied research still has been developed and renovated. The image processing field where wavelet transform is applied almost includes the whole image processing aspects. Video retrieval has been researched in the paper. Based on the analysis in these research actualities, some algorithms and ideas have been proposed and the relevant test result and analysis have been given. The main research contents are as follows.Video sequences are viewed as 2-D among the frames and 1-D of time, so image retrieval algorithms presented by others can be applied to video retrieval. First, the video libraries are filtered by image retrieval algorithm based on wavelet, and then calculate the similarity between query clip and target clip according to the similarity measure principle, the maximal similarity indicates that the target clip is the same as the query clip. The target clip is the one clip to be inquired. Experimental indicates that the proposed methods is not only efficient in retrieving the video clip, but also can gain the higher search speed.From the other aspects, the video data can also be regarded as three -dimensional signal, and is decomposed by 3-D Wavelet Transform. This way overcomes the relations between temporal and spatial. First, the target video library is filtered by the average of wavelet -coefficient standard deviation of the approximation sub-image. Then, the detail feature is extracted using the octree algorithm. Finally, the video clip was retrieved accurately by combining the detail features with the approximate features. Experimental results indicated that the proposed approach is efficient in retrieving video clips.更多还原