基于空间金字塔的镜头检测

Shot detection based on spatial pyramid

提出一种HSV颜色直方图与像素的位置空间金字塔结合的方法,该方法既考虑像素的信息,又考虑了像素的位置信息,来进行视频镜头分割。将图像进行一次亚采样,对得到的第一层亚采样图像进行均匀的四分割;对得到的四块小区域分别计算其颜色直方图,获得四个特征向量;将第一层亚采样的图像再进行一次亚采样,得到第二层亚采样图像,并计算其颜色特征向量;将五个特征向量级联,作为图像帧的特征,并计算相邻两个帧的相似度。该算法不仅考虑了像素信息,还考虑了像素位置信息,而且考虑的是全局位置信息和局部位置信息。实验结果显示,该方法较好地权衡了查全率和查准率。

In this paper, it proposes a new method, which takes into account both pixels＇ information and its location, combin- ing the HSV histogram and the spatial pyramid of pixels to detect shots. A sub-sampling image, which is the first layer of the sub-sampling pyramid, is evenly split to four blocks. It is to extract color features from the four blocks respectively. Making another sub-sampling is to get a second layer image, and it extracts the color feature. It is to connect five vectors to be this frame image＇ s color feature, and calculate the similarity of adjacent frames. This algorithm considers not only the pixel information, but also the location information both of the global and local ones. Experimental results show that this method gives a better bal- ance about recall and precision rates.