期刊论文

Feng, H.(fenghc@whpu.edu.cn)

中文

华中科技大学学报(自然科学版)

1671-4512

2011

39

7

59-62

10.13245/j.hust.2011.07.016

(60873030):国家自然科学基金资助项目 (D20101703):湖北省教育厅重点科研项目 (2010CDB06603;2009Chb008):湖北省自然科学基金资助项目

本校为第一且通讯机构

为了有效降低闪光、字幕、镜头移动等因素对镜头检测的影响,提出了改进的直方图分块法,以各块的帧间差作为特征,以分块权值为特征权重进行模糊聚类,根据视频突变、渐变边界的特点进行预处理,排除了闪光、字幕、镜头移动等对镜头边界检测的影响,从而增强镜头边界检测的鲁棒性,最终实现镜头的边界检测.实验表明该方法对不同类型的视频具有较强的鲁棒性.

To effectively reduce the flash, subtitles, camera movements, and other factors when detecting lenses, the histogram binning was improved. Taking the frame difference of each block as a feature value and taking the feature blocks as the right weight, fuzzy clustering were done. According to the impact of the shot detection, such as video mutation, gradual boundary pretreatment characteristics, remove flash, subtitles, camera movement, shot boundary detection was realized ultimately. Experiments show that the method has strong...