Inspired by the thermodynamic aggregation motion of inhomogeneous Potts model as data clustering,a new segmentation method based on superparamagnetic clustering is proposed for segmenting complicated infrared image.First,the Hamiltonian function for controlling system action is determined.Then,the system’s phase is recognized by measuring the curve of susceptibility vs temperature.Finally,the image is segmented into sub-clusters by measuring the spin-spin correlation function in the superparamagnetic phase.Combined the SW algorithm and Metropolis algorithm,a new method for generating Markov process is proposed,which can converged into Boltzmann distribution quickly,so reducing the computation time of superparamagnetic clustering.The experimental results for complicated infrared images show that the proposed method is obviously better than the SW algorithm in the aspect of convergence speed and segmentation effects.