欢迎访问《图学学报》 分享到：
图学学报

图学学报

• 视觉与图像 • 上一篇    下一篇

基于贝叶斯模型的内容保持图像缩放算法

  

  1. 1. 洛阳师范学院信息技术学院，河南 洛阳 471934；
    2. 洛阳理工学院电气工程与自动化学院，河南 洛阳 471013；
    3. 上海师范大学信息与机电工程学院，上海 200234
  • 出版日期:2017-06-30 发布日期:2017-07-06
  • 基金资助:
    国家自然科学基金项目(U1304616，61502220)；河南省科技攻关计划项目(1721023106361)

Content-Aware Image Resizing Based on Bayesian Model

  1. 1. College of Computer and Information, Luoyang Normal University, Luoyang Henan 471934, China;
    2. College of electrical Engineering and Automation, Luoyang Institute of Science and Technology, Luoyang Henan 471013, China;
    3. The College of Information, Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 200234, China
  • Online:2017-06-30 Published:2017-07-06

PDF (PC)

213

被引次数

1

摘要： 为解决图像在不同显示设备上进行缩放时显著目标易变形、微小目标易删除和多显
著目标易融合等问题，提出一种基于贝叶斯模型的内容保持图像缩放算法。首先,用凸包和背景
先验共同获得贝叶斯模型所需的先验概率和似然估计,代入贝叶斯模型算出显著图；其次，将梯
度图与显著图相乘获得新梯度图，通过求新梯度图和显著图之和获得复合能量图；最后，利用该
复合能量图进行缝缩放。实验结果表明，该缩放算法与原缝缩放算法相比解决了显著目标易变形
和微小目标易删除的问题，减少了多显著目标易融合的现象。

关键词: 内容保持, 显著性检测, 贝叶斯模型, 背景先验

Abstract: In order to solve the problems that salient objects easy to be deformed, small objects easy
to be deleted and multi-salient objects easy to be fused as the image resize in different display devices,
this paper presents a new content-aware image resizing algorithm based on Bayesian model. The
algorithm firstly uses the convex hull and the background prior to obtain the prior probability and the
likelihood estimation required by Bayesian model, and calculate the saliency map using the Bayesian
model. Secondly, after the new gradient map is obtained by multiplying the gradient map and the
saliency map, the new gradient map and the saliency map gets a composite energy map. Finally, we use
the composite energy map to resize the map by seam carving. The experimental results show that the
algorithm compared with the previous algorithm can overcome problems of salient objects deformation
and small objects, and reduces the happening of the multi objects fusion significantly.

Key words: content-aware, saliency detection, Bayesian model, prior distribution