In this thesis, we designed an optimal reference image based on the JTC along with simulated annealing algorithm to recognize the target that has different distortions. We also elaborate on the mathematical model and specific methods. Based on the basic simulated annealing concept and JTC system, we used seven randomized and reference functions with different levels to get the optimal reference image from the training images that have different degrees of rotation. Numerical results have showed that the optimal filter design performs well in image recognition.