Abstract: An image restoration algorithmwas studied to detect tight butt weldbased on magneto-optical imaging technique. A total variation (referred to as ROF) image restoration model was established with energy functional. The problems of weld magneto-optical image degradation and low definition had been solved. The existence of image restoration model solution was demonstrated using the comparison principle of partial differential equations, in which image restoration process was transformed into a constraint optimization problem. Euler-Lagrange equations of restoration model were presented with the functional variation, whose numerical approximate solution of discrete form can be used for image restoration. Tests were carried out to reduce the image noises and restore the images using the total variation restoration model. The edge and texture characteristics of a magneto-optical image could be kept with satisfaction. Experimental results show that the quality of weld magneto-optical image can be promoted by total variation image restoration, and the weld position can be detected accurately.