欢迎访问《图学学报》 分享到:

图学学报

• 图像处理与计算机视觉 • 上一篇    下一篇

基于三角函数迭代的视频数据特征提取

  

  1. (大连大学信息工程学院,辽宁大连 116622)
  • 出版日期:2020-08-31 发布日期:2020-08-22

Feature extraction of video data based on trigonometric function iteration

  1. (College of Information, Dalian University, Dalian Liaoning 116622, China)
  • Online:2020-08-31 Published:2020-08-22

摘要: 在计算机视觉研究中,基于视频数据进行图像对象识别逐渐增多。针对视频数据
的特征提取,提出了一种基于三角函数迭代提取视频三维迭代轨迹特征的方法。该方法是考虑
视频数据的时间空间维度,利用三角函数构造三维动力系统,整体一次性进行视频段数据特征
的抽取,提取出一种近似混沌吸引子的三维特征点集,这种视频数据迭代特征实际上是迭代轨
迹点集合。以VidTIMIT 数据集进行人脸识别实验,发现增加初始迭代值的个数,减少迭代次
数后,提取出的特征点集合具有更好的效果。当VidTIMIT 的43 组559 个视频全部参与实验,
识别率达到88.16%,与现有文献中记载的其他方法相比,具有识别率高、计算时间少的特点,
初步证实了该三维视频迭代轨迹特征具有实用性,同时也值得进一步研究验证与分析。

关键词: 动力系统, 迭代, 视频, 人脸识别

Abstract: In the research of computer vision, the recognition of image objects based on video data is
on an increasing trend. Focusing on the feature extraction of video data, a method based on
trigonometric function iteration was proposed to extract 3D iterative trajectory features of the video.
Considering the time and space dimensions of video data, this paper constructed a three-dimensional
dynamic system by using a trigonometric function, obtained the features of video segment data as a
whole in one extraction, and extracted a set of three-dimensional feature points similar to chaotic
attractors. This iterative feature of video data is an iterative set of track points. Face recognition
experiments using VidTIMIT datasets of face videos show that increasing the number of initial
iterations and reducing the number of iterations could lead to a better effect of the extracted feature
points set. After 43 groups of 559 videos of VidTIMIT were all experimented with, the recognition
rate could reach 88.16%. Compared with other methods recorded in the existing literature, the method
proposed in this paper is characterized by high recognition rate and short computing time. It is proved
that this 3D video iterative trajectory feature is of great practical significance and requires further
research, analysis and verification.

Key words: dynamic system, iteration, video, face recognition