一种基于TV模型结合MRF的图像修复算法

AN IMAGE INPAINTING ALOGRITHM BASED ON TV MODEL AND MRF

在图像修复的算法模型中,全变分TV(total variation)模型对于结构性强的图像拥有较好的修复效果,但对于图像纹理和边缘细节部分修复效果较不理想。而马尔可夫随机场(MRF)下FOE模型采用的邻域相关的系统方法在图像细节纹理修复方面有着出色的表现。故提出一种将两种模型结合的TV-FOE模型用于图像的修复,通过引进混合比例参数,使得改进的模型既能在保证对图像结构层次的修复效果,又能对图像纹理方面拥有很好的复原效果。将其与分别采用TV模型和FOE模型以及其他修复效果较好的算法进行对比,采用客观量化指标峰值信噪比(PSNR)、均方差(MSE)和像素差别图像进行分析、比较,量化结果证明所提出的TV-FOE模型对于破损图像拥有精度更高的修复效果。

In the algorithm models of image inpainting,total variation(TV)model has a good effect on the image with strong structure,but it is not ideal for the image texture and edge details.The neighborhood correlation system method adopted by FOE model in Markov random field(MRF)has excellent performance on image detail texture inpainting.We propose the TV-FOE model which combines the above two models for image inpainting.By introducing the mixed scale parameter,the improved model could not only guarantee the effect of repairing the image structure level,but also have a good restoration effect on the image texture.Compared with TV model,FOE model and other algorithms with good inpainting effect,it was evaluated by objective quantitative indexes:PSNR,MSE and difference image.The quantitative results verify that the proposed TV-FOE model has higher accuracy on inpainting damaged images.