For the thermal images obtained with scan, it happens frequently that the interframe micro-displacement in vertical direction is no longer equivalent to accurate half-pixel. An approximating recursive technique, dealing with the dislocation at different direction, is introduced to reconstruct the image with high resolution. By analyzing the error distribution caused by the approximating recursion, a sub-pixel processing algorithm to reduce the error is proposed. The algorithm uses neighboring pixel to approximate the boundary, and the statistical principle is used to minimize the error introduced by the approximation. The obtained sub-pixel image is highly similar to the original image. All the simulation results and experiment data show that the performance of the algorithm proposed in this paper is obviously better than that of over-sampling and approximating recursive reconstruction. The algorithm is simple and effective. A real time processing is easily achieved. The realization of the algorithm has great significance in improving the technological performance of scanning thermal imaging system.