Journal of Computer Applications ›› 2015, Vol. 35 ›› Issue (10): 2955-2958.DOI: 10.11772/j.issn.1001-9081.2015.10.2955

Previous Articles     Next Articles

Image completion algorithm based on depth information

HE Ye, LI Guangyao, XIAO Mang, XIE Li, PENG Lei, TANG Ke   

  1. College of Electronics and Information Engineering, Tongji University, Shanghai 201804, China
  • Received:2015-05-13 Revised:2015-07-06 Online:2015-10-10 Published:2015-10-14

基于深度信息的图像修复算法

何埜, 李光耀, 肖莽, 谢力, 彭磊, 唐可   

  1. 同济大学 电子与信息工程学院, 上海 201804
  • 通讯作者: 李光耀(1965-),男,安徽安庆人,教授,博士,主要研究方向:图形图像学、虚拟现实,saiheye2009@163.com
  • 作者简介:何埜(1992-),男,安徽安庆人,硕士研究生,主要研究方向:数字图像处理、机器学习;肖莽(1982-),男,江西吉安人,博士研究生,主要研究方向:数字图像处理、机器学习;谢力(1992-),男,浙江台州人,博士研究生,主要研究方向:数字图像处理、机器学习;彭磊(1977-),男,山东泰安人,博士研究生,主要研究方向:数字图像处理;唐可(1991-),男,重庆人,硕士研究生,主要研究方向:数字图像处理。
  • 基金资助:
    国家自然科学基金资助项目(60771065,51378365)。

Abstract: Aiming at the problem of object structure discontinuity and incompleteness occurred in image completion, an image completion algorithm based on depth information was proposed. Firstly, the plane parameter Markov random field model was established to speculate depth information of the pixels in the image where the scene situate, then the coplanar region in the image determined, and the target matching blocks were located. Secondly, according to the principle of perspective projection, the transformation matrix was derived, which guided the geometric transformation of the matching blocks. Finally, the target cost function which includes the depth term was designed. Experimental results show the proposed algorithm has superiority in both subjective details and Peak Signal-to-Noise Ratio (PNSR) statistics.

Key words: image completion, depth information, perspective projection, global optimization

摘要: 针对图像修复结果中存在物体结构上的不连续性和不完整性的问题,提出一种基于深度信息的图像修复算法。首先,通过建立平面参数马尔可夫模型来推测图像中像素点所在场景空间中的深度信息,从而划分出图像中的共面区域,定位匹配块的目标位置;其次,根据透视投影原理推导出相应的变换矩阵,指导对匹配块的几何变换;最后,设计含义深度项的目标代价函数,通过全局优化选择最优修复效果。实验的主观效果对比和峰值信噪比(PSNR)数据都表明该方法更为优越。

关键词: 图像修复, 深度信息, 透视投影, 全局优化

CLC Number: