CAJ | 학술논문

基于旋转不变均衡局部二值模式，提出一种均衡生成速度和信息量的视频算法。首先，使用预采样方法降低视频处理数据量，在此基础之上提取图像的局部二值模式特征；然后对两帧图像相似度进行分析，获取聚类数目。获取聚类数目之后，使用k均值算法对镜头关键帧进行聚类；最后，使用“重要度”函数评测聚类重要度，从“重要”聚类中选取聚类中心最近帧为关键帧。实验结果表明，该算法生成的视频在保证实时性的同时，提高的信息量，较好地表达了视频的内容。
기우선전불변균형국부이치모식，제출일충균형생성속도화신식량적시빈산법。수선，사용예채양방법강저시빈처리수거량，재차기출지상제취도상적국부이치모식특정；연후대량정도상상사도진행분석，획취취류수목。획취취류수목지후，사용k균치산법대경두관건정진행취류；최후，사용“중요도”함수평측취류중요도，종“중요”취류중선취취류중심최근정위관건정。실험결과표명，해산법생성적시빈재보증실시성적동시，제고적신식량，교호지표체료시빈적내용。
Aiming at making an improvement on the information capacity of video summary under the premise of summary speed , a new method for video summarization based on rotation invariant uniform pattern feather is presented in this paper .First of all, the amount of frames to be processed is decreased with the pre-sampling approach , upon which local binary pattern feature is ex-tracted.Then, the cluster number is generated according to the similarity between each two adjacent frames .With the cluster number, the k-means algorithm is applied to cluster similar frames .Finally, the frame which is nearest to the cluster center is used to construct the video summarization .The importance of cluster is evaluated by the function introduced .Experiment results indicate that the proposed method can making an improvement on the information capacity of video summary under the premise of generation speed , the summary performs well to the source video .