计算机工程与应用 ›› 2015, Vol. 51 ›› Issue (16): 161-165.

• 图形图像处理 • 上一篇    下一篇

一种改进的时空线索的视频显著目标检测方法

秦利斌,刘纯平,王朝晖,季  怡   

  1. 苏州大学 计算机科学与技术学院,江苏 苏州 215006
  • 出版日期:2015-08-15 发布日期:2015-08-14

Approach of detecting salient objects in videos using spatiotemporal cues

QIN Libin, LIU Chunping, WANG Zhaohui, JI Yi   

  1. Shool of Computer Science and Technology, Soochow University, Suzhou, Jiangsu 215006, China
  • Online:2015-08-15 Published:2015-08-14

摘要: 针对Zhai和Shah提出的原始时空显著性检测模型在空间显著性方面仅仅使用了图像的亮度信息,忽略彩色图像中的色彩信息的不足,提出了一种基于HSV颜色模型的空间显著性计算方法。该方法充分利用图像中的亮度信息和彩色信息,从像素级和区域级两个层次上进行显著性的计算。将改进的空间显著性计算与Zhai和Shah提出的时间显著性计算以及时空融合框架进行整合,检测视频中的显著目标。实验证明改进方法在光照不均和背景较复杂的情况下获取的空间显著区域和显著目标比原始方法更准确。

关键词: 时间显著模型, 空间显著目标检测模型, 动态融合, 目标检测, 复杂场景

Abstract: The original model of spatiotemporal visual attention detection only uses intensity of image and doesn’t consider the color information in spatial attention model. A novel technique is proposed to get spatial saliency map. The spatial saliency is computed in two levels:pixel level and region level. The proposed technique takes advantage of intensity and color information. The results of experiment demonstrate the proposed technique get more accuracy in obtaining salient region and object detection to visual scene with unevenly illumination and complex background.

Key words: temporal attention model, spatial attention model, dynamic fusion, object detection, complex scene