Abstract: Through analyzing the least squares support vector regression (LS-SVR) filtering characteristics, the LS-SVR method is presented to construct the convolution mask of image filtering, which resolves the solver problem in the LS-SVR application. Based on the method, the LS-SVR switching filter for image corrupted by salt pepper noise is proposed. By taking the maximum-minimum operator as the salt pepper noise detector in the filtering algorithm, using the non-noise point of the filtering windows as the input data set of the LS-SVR, and utilizing the LS-SVR filtering operator to execute simple convolution operation on the filtering windows, the corrupted data are efficiently restored. Experiments show that the proposed algorithm has better detail preserving ability and better noise removing ability.