基于特征知识库的动漫视频暴力场景检测方法

Detection method of animation video violence scene based on feature knowledge base

摘要:

针对海量动漫视频剪辑中可能包含不符合观众认知或接收程度的内容，为观众带来不愉快体验的问题，提出了基于知识库的卡通视频暴力场景检测方法。首先，利用低级特征(RBG颜色、亮度、色调)开发连续函数，进行场景中镜头边界识别；然后，利用运动信息，计算视频帧中的运动。接着，使用空间信息和运动信息进行分割处理，并使用欧氏距离方法进行对象识别；最后，利用包含所有常见物品、卡通角色及其视觉特征和行为特征的知识库，计算场景的暴力概率。实验结果表明，该系统能够成功检测到卡通视频中的暴力场景。

Abstract:

Aiming at the problem that the massive video clips may contain content that does not meet the audience's cognition or reception, which brings unpleasant experience to the audience, a cartoon video violence scene detection method based on knowledge base is proposed. Firstly, the continuous function is developed by using low-level features (RBG color, brightness and hue) to recognize the shot boundary in the scene; Then, the motion information is used to calculate the motion in the video frame; Then, the spatial information and motion information are used for segmentation, and the Euclidean distance method is used for object recognition; Finally, the violence probability of the scene is calculated by using the knowledge base containing all common objects, cartoon characters and their visual and behavioral characteristics. The prototype of the method is developed and implemented by MATLAB. The experimental results show that the system can successfully detect the violence scene in cartoon video.