基于场景分割的视频内容语义管理机制

摘要

To tackle video management problem of semantics gap existing in different aspects,a video semantic de-scription framework based on UCL (Uniform Content Locator)is proposed.The semantic description framework consists of three levels,i.e.,content,control and physical.Video to be semantically managed is divided into different scenes based on spatial-temporal similarities of frames.For every scene,the most optimal reference frame (I-frames)and non-optimal refer-ence frames (non I-frames)are selected based on local texture complexity,background luminance and scene complexity. Content semantic are imbedded into non I-frames while control and physical semantics are imbedded into I-frames.A seman-tic watermarking algorithm is incorporated into the management to realize the efficient storage and transmission of video con-tent and its video semantics.JM reference model is adopted for experiments to verify the watermarking technique and results show that the method is robust and has little side effect on video quality.%针对视频内容管理在不同层面存在语义鸿沟的问题，提出基于UCL（Uniform Content Locater ）的视频语义描述框架，该框架包含了三个层次的语义：内容语义、控制语义以及物理属性信息。而视频场景的分割则通过视频内容基于时空上的相似性实现。对于每个视频场景，结合局部纹理复杂度、背景亮度和场景复杂度，选择最佳参考帧（I帧）与非最佳参考帧（非I帧）以嵌入不同的语义信息：控制语义、物理属性信息嵌入I帧，内容语义嵌入非I帧。利用数字语义水印技术来实现视频内容的语义管理，完成语义信息和载体信号的一体传输和存储。实验中采用JM参考模型进行数字水印方法的验证，结果表明该方法鲁棒性强，且不会造成视频资源质量显著下降。