| With the development of digital devices,the amount of information in modern so ciety is growing fast.As an important information carrier,videos become more and m ore popular in daily life.Under such circumstances,quick and accurate video classific ation is particularly important.The state of art on video classification based on tenso r decomposition usually requires the unified length of video sequence so that the infor mation of video data can’t be fully used.Based on the video sequences as the research object,this dissertation focuses on the problem of unequal length of video sequences.We propose a tensor decomposition algorithm based on the main problem of the unequal length of the video sequences,which can improve the above problems effectively when maintain a high accuracy.The main contents include:(1)Spatial-Temporal Iterative Tensor Decomposition algorithm: starting from the problem of unequal length of video sequences,this paper analyzed the characteristics of different sampling method.After improved it,a method to automatically extract video frames is proposed.This algorithm can output the equal videos preserving the characteristic information to the maximum degree.(2)Video sequence classification framework based on tensor decomposition: after the realization of the equal length of the videos,we unfold the new generated tensor in the third order,and then take each order matrix as a subspace.We make the use of principal angles between subspaces to classify the video sequences and establish the video sequences classification framework based on tensor decomposition.The innovation of this article is combining tensor composition with video sequences classification,which resolves the problem of unequal length of video sequences.At the same time,this paper proposed a video sequences classification framework based on tensor decomposition.The experiments show that compared with other traditional algorithms,the algorithm can acquire a better classification result. |
