基于插值的图像修复算法

Image Inpainting Algorithms Based on Interpolation

目前,针对小区域缺损的图像修复算法中,大多采用基于迭代的修复算法,然而这些效果较好的图像修复算法,其时间复杂度一般都比较大.本文通过插值的方法,利用破损区域与周边邻域的有效信息之间的相关性,提出改进的基于FMM(快速行进)的图像修复算法;调用仅适用于被检测到的高局部活跃的像素,提出基于LMMSE(线性最小均方误差估计)插值的图像修复算法.实验结果表明,提出的两种算法分别与FMM算法和TV(整体变分法)算法相比,在整体上其修复效果和效率都具有明显的优势.

For image repair in a small defect area, iterative algorithms with high computational complexity are commonly used. In this paper, an improved fast marching method （FMM） is proposed by using correlation information in the neighborhood of damaged area. A new image restoration algorithm based on LMMSE interpolation is also proposed by using pixels with high local activity only. Experimental results show that the proposed algorithms have improved performance as compared with the FMM method and integral variation method（ TV ）.