CAJ | 학술논문

关键帧提取是基于内容的视频检索技术的关键问题.文中提出一种基于多重压缩感知和距离计算的关键帧提取算法,首先将镜头内的各帧图像分割为若干不相交的块,通过滤波器生成块的高维特征;然后利用多个不同的、具有有限等距性质的稀疏矩阵对块高维特征进行采样,将采样的均值作为块的低维特征;采用多种距离计算相邻帧对应块之间的差异,完成子镜头的分割操作,在每个子镜头内部,选取与该子镜头平均内容最接近的帧作为关键帧.实验表明,该算法提取出的关键帧能够准确描述镜头的主要内容.
관건정제취시기우내용적시빈검색기술적관건문제.문중제출일충기우다중압축감지화거리계산적관건정제취산법,수선장경두내적각정도상분할위약간불상교적괴,통과려파기생성괴적고유특정;연후이용다개불동적、구유유한등거성질적희소구진대괴고유특정진행채양,장채양적균치작위괴적저유특정;채용다충거리계산상린정대응괴지간적차이,완성자경두적분할조작,재매개자경두내부,선취여해자경두평균내용최접근적정작위관건정.실험표명,해산법제취출적관건정능구준학묘술경두적주요내용.
Key frame extraction is considered as the key issue of content-based video retrieval.An algorithm based on multiple compressive sensing and distances′computing is proposed.In the first step, each frame in one shot is segmented into several disjoint blocks, high dimensional features of which are generated by filtering. Then, multiple different sparse matrices that satisfy restricted isometry property are employed to sample the high dimensional feature of each block, and the mean value of sampling is calculated as the low dimensional feature of each block.Several different distances are used to compute differences between corresponding blocks of neigh-boring frames to conduct sub-shot segmentation.The frame nearest to the average content of each sub-shot is se-lected as the key frame.Experimental results demonstrate that key frames extracted by the proposed algorithm can describe the main content of shot accurately.