Unsupervised ESD Method Based on GMM Clustering in Video Retrieval
QI Xing-bi · Video Engineering · 2015
For the purpose of an efficient retrieval of impressive scenes from videos,an emotional scene detection method based on Gaussian mixture model is proposed. Firstly,42 feature points are selected from facial,and 10 features are defined. Then,Gaussian mixture model is used to divide video into multiple clusters. Finally,emotion scene is divided into single cluster by using facial expression classification results of each frame,and scene integrating and deleting is used to finish detecting. Experimental results on life record video and MMI face expression database show that the detecting and classification rate of proposed method can achieve98% and 95% respectively. It takes only 0. 138 seconds in detecting emotion scene video with five minutes. Proposed method has better performance than several advanced detecting methods.