北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2014, Vol. 37 ›› Issue (3): 27-31.doi: 10.13190/j.jbupt.2014.03.006

• 论文 • 上一篇    下一篇

基于视觉显著性检测的跨尺度插值模型

訾玲玲1,2, 杜军平1, 汪谦1   

  1. 1. 北京邮电大学 智能通信软件与多媒体北京市重点实验室, 北京 100876;
    2. 辽宁工程技术大学 电子与信息工程学院, 辽宁 葫芦岛 125105
  • 收稿日期:2013-09-23 出版日期:2014-06-28 发布日期:2014-06-08
  • 作者简介:訾玲玲(1981-),女,博士生,E-mail:lingling19812004@126.com;杜军平(1963-),女,教授,博士生导师.
  • 基金资助:

    国家重点基础研究发展计划项目(2012CB821206);国家自然科学基金项目(61320106006)

Transcale Interpolation Model Based on Visual Saliency Detection

ZI Ling-ling1,2, DU Jun-ping1, WANG Qian1   

  1. 1. Beijing Key Laboratory of Intelligent Telecommunication Software and Multimedia, Beijing University of Posts and Telecommunications, Beijing 100876, China;
    2. School of Electronic and Information Engineering, Liaoning Technical University, Liaoning Huludao 125105, China
  • Received:2013-09-23 Online:2014-06-28 Published:2014-06-08

摘要:

为了提升多媒体处理感知能力,提出了一种运动图像序列跨尺度插值模型. 首先从用户的角度出发,采用视觉显著性检测技术获取关注区域,进而确定运动图像的显著区域;而后在时间尺度改变的情况下,利用一致敏感哈希技术实现了高质量插值帧的生成;最后定义能量函数,使得在分辨率尺度变化时能提升关注区域的清晰度. 实验结果表明,运动图像序列跨尺度插值模型能够获得较好的插值质量.

关键词: 跨尺度, 插值模型, 视觉显著性检测, 图像序列

Abstract:

In order to improve the perception ability of multimedia processing, a transcale interpolation model is proposed for the dynamic image sequences. Firstly, the visual attention regions are obtained by utilizing visual saliency detection from the viewpoint of users, so as to determine the outstanding regions of dynamic image sequences. Secondly, coherency sensitive hashing is adopted and high quality interpolation frames are generated when the time scale is changed. Finally, the energy function of the target images is defined, so the quality of visual attention regions could be improved when the resolution scale is changed. Experiments show that the proposed model achieves better interpolation results.

Key words: transcale, interpolation model, visual saliency detection, image sequences

中图分类号: