【作者】 刘飞；

【导师】 罗代升；

【作者基本信息】 四川大学 ， 通信与信息系统， 2005， 硕士

【摘要】 随着视频数据量的不断扩大,不仅要求数据库和其它信息系统能对视频进行存储和基于关键字的检索,而且要对多媒体数据的内容进行语义分析,以达到更深的检索层次,从而实现基于内容的视频检索。由于基于内容的视频检索的数据量非常大,检索速度慢,因此本文的目的是压缩视频检索的数据量,提高检索速度。 本文提出两种方法压缩视频检索的数据量,提高检索速度。一种方法是改进的HSV量化方法,降低色彩信息量。另一种方法是采用镜头分割,提取代表帧,减少搜索帧数。 改进的HSV量化方法是针对颜色特征,把纯颜色非均匀量化为六个等级,对于饱和度和亮度,都非均匀量化成三个等级,最后共得到27个颜色分级。试验结果表明,采用本文的量化方式计算量小,图像质量较好。由于对亮度和饱和度不敏感,减少了颜色偏移的可能性。同时,由于量化时采用的是符合人类视觉特征的非均匀量化,在基于内容的检索时会检索到用户感兴趣的颜色特征。 镜头分割方法是采用基于直方图方向差异的分割方法,即先计算出相邻两帧基于HSV的直方图,再将两帧的直方图作为参数代入到方向差异函数中,得到最后的距离值。镜头分割方法用于代表帧的提取。对比现有几种方法的分割结果表明,本算法实现简单,镜头分割的准确率较高。 选取合适的代表帧能够减少搜索帧数,即减少视频搜索数据量。本文使用了基于方向积累差异的代表帧选取方法。使用方向积累差异和镜头分割两个条件作为代表帧选取的准则。只有同时满足这两个条件,视频中的一帧才能被选更多还原

【Abstract】 With the enlargement of the data quantity of the video, not merely require the database and other information systems to carry on storing and retrieval based on key word to the video, and should carry on semantic analysis to the content of the multimedia data, in order to reach deeper retrieval level, thus realize content-based video retrieval. Because the amount of analyzed video data is enormous, we need reduce it. This paper carried on the work around the reduce of video data and enhance the retrieval speed.This paper proposed two methods to compress the video data, enhance the retrieval data. The first one is an improved HSV quantization method, to reduce the colors. The other method need detect the shot boundary, and then select key frame. The purpose of this method is to reduce the frames need to be searched.The improved HSV quantization method is to the color feature. This paper has non-uniform quantize pure color to six colors, to saturation and luminance, non-uniform quantize to three grades. Finally , there are 27 colors altogether. The result of the experiment indicates, the quantization method we proposed computational cost is low; The quality of the image is better; Because not sensitive to the luminance and saturation, reduce the possibility of color skew. Meanwhile,because it was the non-uniform quantization which accords with the human vision characteristic, it contributes to finding the color features of user’s interest.This paper proposed a shot boundary detection method that based on HSV histogram and directed Divergence. This method calculates out HSV histograms of two frames, then regard the histograms of two frames as the parameter of directed divergence function, return the last distance value. This algorithm is easy to be realized, Accuracy is relatively high to compare with the results of other methods.A good key frames select method can reduce the frames need to be searched, enhance the speed of retrieval. This paper used a cumulative directed divergence method choose key frames. Cumulative directed divergence and shot boundary affect the select of key frames. Meeting two conditions only at the same time, it could be elected as key frames of the video. Realization of this algorithm is very simple, the chosen key frames is more representative.更多还原