【摘要】 针对视频中的渐变镜头边界难以检测的问题,提出一种基于循环神经网络的渐变镜头检测方法。使用深度神经网络inception-v3提取图像帧的特征并计算帧间相似度,根据相似度序列的特点初步找出候选渐变片段;以帧间差为输入向量训练一种对视频片段中视频帧的类型分类的循环神经网络模型,通过网络模型对候选渐变片段的帧分类,找出准确的渐变镜头边界。在TREC2001视频数据集上与其它渐变镜头检测方法进行对比,实验结果表明,该方法具有较高的准确性。更多还原

【Abstract】 Aiming at the problem that the gradient shot boundary in video is difficult to detect,agradual shot boundary detection method based on recurrent neural network(RNN)was proposed.The deep neural network inception-v3 was used to extract the features of the image frame and calculate the similarity between frames.The candidate gradient segments were initially found according to the characteristics of the similarity sequence.A recurrent neural network model for classifying the video frames in the video segment was trained with the frame difference as the input vector,and the frame of the candidate gradient segments was classified by the network model to find the exact gradient lens boundary.Compared with other gradient shot boundary detection methods on the TREC2001 video dataset,experimental results show that the proposed method has high accuracy.更多还原