欢迎访问《图学学报》 分享到：
图学学报

图学学报

• 专论：第16届媒体智能与大数据计算会议（CIDE & DEA 2019 大连） • 上一篇    下一篇

一种基于视频数据的服装显著性预测方法

  

  1. (1. 华北电力大学控制与计算机工程学院，北京 102206； 
    2. 中国科学院计算技术研究所，北京 100190)
  • 出版日期:2019-12-31 发布日期:2020-01-20
  • 基金资助:
    国家自然科学基金面上项目(61972379)；国家自然科学基金青年基金项目(61300131)

A Clothing Saliency Prediction Method Based on Video Data

  1. (1. School of Control and Computer Engineering, North China Electric Power University, Beijing 102206, China; 
    2. Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China)
  • Online:2019-12-31 Published:2020-01-20

PDF (PC)

109

摘要： 人眼视觉注意机制表明当人眼观察目标时，注意力只会放在少数感兴趣的区域， 而自动忽略视野中大部分不感兴趣的其他区域。研究人类视觉注意机制，并构建有效的服装显 著性预测模型，可在后期用于指导更加逼真有效的服装运动建模，提高模拟效率。为此，对着 装人体运动视频数据进行分析，构造了种类多样的视频样本，并利用眼动技术采集真实人眼的 注视数据，采用高斯卷积生成视频帧的显著图作为训练模型所需的 Ground-truth。在进行视频特 征提取时，结合了底层图像特征、高层语义特征以及运动特征，共同构造特征向量和标签，并 通过支持向量机(SVM)训练得到基于服装视频的显著性预测模型。通过实验验证，该方法的性 能在服装显著性预测时，优于传统的显著性预测算法，具有一定的鲁棒性。

关键词: 视觉注意机制, 服装建模, 眼动技术, SVM, 显著性预测

Abstract:

The human visual attention mechanism shows that when the human eyes look at the target, the attention will only be focused on few areas of interest, while most of the other areas out of interest in the field of vision will be automatically ignored. The study of human visual attention mechanism and the construction of an effective clothing saliency prediction model can be used to guide more realistic and effective clothing motion modeling and improve the efficiency of simulation. In this paper, we analyzed the video data of the dressed human movement, constructed a variety of video samples, and adopted eye movement technology to collect the gaze data of real human eyes. Gauss convolution was used to generate the salient image of video frame as the Ground-truth required for training model. In the video feature extraction, the underlying image features, high-level semantic features and motion features were combined to construct feature vectors and tags, and the significance prediction model based on clothing video was obtained by support vector machine (SVM) training. The experimental results show that the proposed method outperforms the traditional significance prediction algorithm and hassome robustness in clothing saliency prediction.

Key words:  visual attention mechanism, clothing modeling, eye movement technique, SVM, saliency prediction