Abstract: To solve the problems of traditional variation inpainting algorithms, like insufficient
information utilization, destroyed texture and strong artificial interference, an adaptive high order
variational image inpanting algorithm based on eight neighbors is presented. Firstly, it makes full use
of eight neighbors at the points to be repaired in damaged images and divides them into two groups of
four-neighbor, then repairs images by nonlinear anisotropic diffusion in each four-neighbor
respectively. The optimal value of parameter p in every four-neighbor will be determined by adaptive
method. Then those points will be discreted through the central difference method. Finally, the
Gauss-Jacobi Iteration of the damaged images can be obtained by using weighting and averaging.
Compared with the image inpainting algorithms in recent years, the simulation results show that, the
images obtained by the proposed algorithm have good evaluations. Not only is the repair time
reasonable, but the texture structure is closest to the original images and values of PSNR or SA are
the highest.

Key words: neighbors, adaptive, high order variation, central differencing, image evaluation