The traditional two-state hidden Markov model divides the high frequency coefficients only into two states (large and small states). Such scheme is prone to produce an inaccurate statistical model for the high frequency subband and reduces the quality of fusion result. In this paper, a fine-grained multi-state contextual hidden Markov model (MCHMM) is proposed for infrared and visible image fusion in the non-subsampled Shearlet domain, which takes full consideration of the strong correlations and level of details of NSST coefficients. To this end, an accurate soft context variable is designed correspondingly from the perspective of context correlation. Then, the statistical features provided by MCHMM are utilized for the fusion of high frequency subbands. To ensure the visual quality, a fusion strategy based on the difference in regional energy is proposed as well for lowfrequency subbands. Experimental results demonstrate that the proposed method can achieve a superior performance compared with other fusion methods in both subjective and objective aspects.