基于多特征的视频镜头检测方法

Video shot boundary detection algorithm based on multi-features

针对视频镜头边缘检测准确率低的问题,提出了一种新的基于多特征的视频镜头检测算法。首先按时序读取多帧图像,并转换为灰度图;进一步将帧图像均匀分块,计算每个图像块的平均梯度,构造视频动态纹理;比较相邻帧视频动态纹理的相关性及两帧SIFT特征的匹配程度,根据匹配结果得出预检测结果;接下来与步长低于人眼刷新频率的下一帧动态纹理及SIFT特征相比较,得到最终的结果。通过对多组不同类型的视频数据进行实验,均能取得较高的召回率和准确率。该文算法对结构较复杂的渐变镜头进行检测,也能取得较高的检测准确率和召回率。

Aiming at the problem of low accuracy in video shot boundary detection, a new video shot boundary detection algorithm based on multi-features is put forward. At first, the multiple frame images in a video are read sequentially and converted to gray images. Then, these frame images are segmented into image blocks evenly, the average gradient of each image block is calculated to form video dynamic texture. The correlation of the video dynamic texture among adjacent frames and the matching degree of the SIFT features of two frames are compared; and the preliminary detection result is obtained according to the matching result. Next, the dynamic texture and SIFT feature of the original frame are compared with those of the next frame with the step distance lower than the refresh rate of the human eyes, and the final detection result is achieved. The experiments on multiple groups of different types of video data were conducted, experiment results show that the proposed algorithm can achieve higher precision and recall rate than some other algorithms. For the gradual changed videos with complex structure, the proposed algorithm can also achieve good results.