摘要： 为快速稳健地检测镜头边界，提出基于互信息量的镜头边界检测算法。采用基于HSV空间不均匀分块直方图求取的帧间互信息量作为视频帧间差异度量，结合相应的阈值策略与时域窗策略，实现对镜头的切变，以及常见渐变与计算机特效技术制作的镜头变换检测。对包括广告、综艺节目、新闻等多类视频数据进行实验，结果证明，该算法对切变检测的综合效果比经典基于直方图的双阈值法平均提高11.9%，渐变检测的综合效果平均提高7.6%，能有效减弱摄像机运动与光照变化对镜头检测的影响，对不同类型视频镜头检测效果稳定，具有较高的查全率与查准率。
关键词:
镜头边界检测,
镜头突变,
镜头渐变,
互信息量,
淡入淡出,
帧间差异
Abstract: Aiming at improving the precision of shot boundary detection, a new algorithm based on mutual information is proposed. The algorithm adopts the mutual information, which is calculated based on non-uniform block histogram in HSV space, as the difference measurement between frames. Combined with the corresponding threshold strategy and the time’s window strategy, the cut and gradual change shots and shot boundaries made by computer effects technology can be successfully detected. Experiments conducted on many kinds of videos such as advertising, variety shows and news video show that, compared with the original twin-threshold method, the proposed algorithm enhances the detection results by 11.9% in cut shots and 7.6% in gradual shots. The algorithm is robust for camera movement and the light changes on shot detection, with higher recall rate and precision rate.
Key words:
shot boundary detection,
shot cut,
shot gradual change,
mutual information,
fade in and fade out,
inter-frame difference
中图分类号: