Abstract: The various total variation (TV) algorithms currently reconstruct the lost or deteriorated parts of images by related information of damaged pixel and its 4 neighborhood pixels. Their inpainting accuracy is low because of finite related information. The double cross TV algorithm proposed in this paper divides 8 neighborhood pixels into 2 groups, and computes the pixel value of damaged pixel by using 4 neighborhood pixels in each group respectively. Therefore, the weighted mean of these two pixel values is the final inpainting pixel value. The Experiments show an improvement in PSNR and less iteration number compared with the original TV.