视频底层特征选取及其与观众评价的相关分析

发布时间：2018-03-18 04:18

本文选题：互相关特征选择方法　切入点：CFS-Spearman　出处：《东北电力大学》2016年硕士论文　论文类型：学位论文

【摘要】：近年来,随着信息技术的迅猛发展,在人们的日常生活中各种各样的视频信息随处可见,目前的电影预告片、广告等视频经常在一些在线社交网络(Online Social Networks,OSN)中预先投放。在视频的设计和生产过程中,怎样设计视频更能够吸引观众的注意,对观众的情绪产生正面的影响,从而提高此类视频的观众口碑和销售业绩,产生经济效益,这是相关文化企业、视频指导者、编辑者非常关心的问题。为探究商业视频底层视觉特征对其受欢迎程度的影响情况,需要对视频底层特征进行提取,在获取原始视频数据及其观众评价之后,采用特征选择方法提取对评价影响较大的特征。本文针对目前广泛应用的互相关特征选择(Correlation-Based Feature Selection,CFS)方法进行改进,提出了CFS-Spearman特征选择算法,采用Spearman秩相关系数与CFS方法结合进行特征选择。选取UCI机器学习数据库中的3组数据作为实验数据源,将本文特征选择方法与传统的CFS方法及最小冗余最大相关(minimum redundancy and maximum relevance,mRMR)方法进行比较,通过KNN、SVM分类器验证各候选集的优劣,证明了文中方法的效果。最终将本文的方法应用于商业视频底层特征选择,选出了对视频受欢迎程度影响最大的特征子集。再用KNN、SVM方法对特征进行分类及验证性实验,结果证明选出的运动方差、亮度均值、饱和度均值三个视频底层特征会对观众的评价造成一定的影响,至此预测模型已建成。通过ROC曲线验证了CFS-Spearman方法与SVM结合优于CFS-Spearman方法与KNN的结合,从实验结果看出CFS-Spearman方法与SVM结合更适合此方面的研究,这一研究成果为以后更进一步研究奠定了基础。
[Abstract]:In recent years, with the rapid development of information technology, all kinds of video information can be seen everywhere in people's daily life. Videos such as ads are often preloaded on online social networks online Social Networks. How to design videos in the process of designing and producing them can attract the attention of viewers and have a positive impact on the mood of the audience. To improve the audience reputation and sales performance of such videos, and generate economic benefits, this is the relevant cultural enterprises, video directors, To explore the impact of the underlying visual features on the popularity of commercial video, it is necessary to extract the underlying features of the video, after obtaining the raw video data and their audience evaluation, The feature selection method is used to extract the features that have a great impact on the evaluation. In this paper, the Correlation-Based Feature selection (CFS) method, which is widely used in cross-correlation feature selection, is improved, and a CFS-Spearman feature selection algorithm is proposed. Spearman rank correlation coefficient and CFS method are used to select the features. Three groups of data in UCI machine learning database are selected as experimental data sources. The method of feature selection is compared with the traditional CFS method and the minimum redundancy and maximum MRs method, and the KNN-SVM classifier is used to verify the advantages and disadvantages of each candidate set. The effectiveness of the proposed method is proved. Finally, the proposed method is applied to the bottom feature selection of commercial video, and the feature subset which has the greatest influence on the popularity of video is selected. The results show that the three underlying video features, such as motion variance, luminance mean and saturation, will have a certain impact on the audience evaluation. The ROC curve shows that the combination of CFS-Spearman and SVM is better than that of CFS-Spearman and KNN. The experimental results show that the combination of CFS-Spearman and SVM is more suitable for this research. The research results lay a foundation for further research.
【学位授予单位】：东北电力大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】