基于在线Bayesian决策的动画场景切换检测方法

动画视频分析中，实时在线地检测场景切换点是一个基础任务。传统基于像素和阈值的检测方法，不仅需要存储整个动画视频，同时检测结果受目标运动和噪声的影响较大，且阈值设定也不太适用复杂的场景变换。提出一种基于在线Bayesian决策的动画场景切换检测方法，新方法首先对动画帧图像分块并提取其HSV颜色特征，然后将连续帧的相似度存入一个固定长度的缓存队列中，最后基于动态Bayesian决策判定是否有场景切换。多类动画视频的对比实验结果表明，新方法能够在线且更稳健地检测出动画场景切换。

Fast animation scene detection using online Bayesian decision

Animation scene detection is a fundamental task for many advanced animation video analysis. Traditional detection methods that are based on pixels’ analysis and threshold, may require huge storage and not be robust to noises and normal object motions. Additionally, it is also difficult to define suitable thresholds for complicated animation scenes. A new online animation scene detection method is proposed in this paper. At first, the proposed method segments video frame images into some sub-blocks and calculate their HSV feature values. Then the distance between two consecutive video frames are recorded into a buffer queue with fixed length. Finally, the proposed dynamic Bayesian decision rules are used to determine whether there exists a scene change. The experimental results on several types of real video demonstrate that the new method can gain better performance on detection accuracy and computation efficiency.