欢迎访问《图学学报》 分享到:

图学学报

• 视觉与图像 • 上一篇    下一篇

基于平行坐标下降法的图像修复

  

  • 出版日期:2015-04-30 发布日期:2015-06-03

Image Inpainting Based on Parallel Coordinate Descent Method

  • Online:2015-04-30 Published:2015-06-03

摘要: 以压缩传感和稀疏表示为理论依据,提出了一种基于平行坐标下降法的图像修复
模型。该模型用小波变换作为图像的稀疏表示,以稀疏性作为正则化项;同时基于松弛阈值来
标记函数实现全局优化,并采用该模型算法得到全局最优解。从峰值信噪比、收敛速度和视觉
效果等3 个方面验证了算法的有效性。结果表明新的模型无论是在客观还是视觉主观上都有更
好的效果,同时算法具有更快的收敛速度。

关键词: 稀疏表示, 图像修复, 平行坐标下降法, 阈值

Abstract: Based on the compressed sensing and sparse representation theory, and on the parallel
coordinate descent an image inpainting model method is proposed, which uses the method of wavelet
transform for image sparse representation, with sparse as the regularization term, and at the same time,
realizes global optimization based on the threshold of relaxation to mark function. The global optimal
solution can be obtained by using PCD. The proposed algorithm is verified from the peak signal to
noise ratio of convergence speed and visual effect. The results show that the new model has better
effect in both objective and subjective, and at the same time has faster convergence speed.

Key words: sparse representation, image inpainting, parallel coordinate descent method, threshold