| 摘要: |
| 目的 针对传统视频镜头边界检测算法精度低和较复杂等缺点， 提出一种基于谱残差显著图和分块灰度直方图相结合计算似然比的算法。方法 为了提取有效结构信息作为特征来比较帧间差异，采用谱残差方法得到各帧图片的显著图。将显著图分块并计算各块的灰度直方图数据， 进一步提高检测的精度。计算各帧似然率后用文中提出的帧间差指标计算比较帧间差异。结果 所提方法的综合检测精度能达到93.82%， 对常见影响因素下相邻未变化帧的相似性保持在99%以上。结论 实验结果表明， 文中算法的检测精度高于传统方法， 过程相对简单， 对常见影响因素具有较强的鲁棒性。 |
| 关键词: 镜头边界检测 谱残差 显著图 分块直方图 似然比 |
| DOI： |
| 分类号:TS807 |
| 基金项目: |
|
| Video Shot Boundary Detection Based on Spectral Residual Saliency Map |
|
KUANG Sheng-kun, WANG Xiao-hong
|
|
University of Shanghai for Science and Technology， Shanghai 200093， China
|
| Abstract: |
| In order to overcome shortcomings such as the low accuracy and the high complexity of the traditional approach of video shot boundary detection, an algorithm using spectral residual saliency map combined with sub-block gray histogram to calculate likehood ratio was proposed in this paper. First, to extract effective structure information as the feature to compare differences between frames, the spectral residual approach was used to obtain the structure saliency map of frames. Second, the frames were segmented into necessary subblocks to calculate all the histogram data which further improved the precision. The likehood radio of each frame was calculated then the frame-to-frame difference index proposed in this paper was computed to calculate and compare frame-to-frame differences. The detection precision of the proposed method can reach 93.82%, and the similarity of unchanged neighboring frames maintained as more than 99%. Experimental results showed that the method was more effective which had a higher precision and relatively simpler process， and it had a strong robustness against common influencing factors. |
| Key words: shot boundary detection spectral residual saliency map sub-block histogram likehood radio |
