This paper proposes an adaptive Markov Random Field（ aMRF） model for the dense matching of deep space exploration images

which usually lack textures. Compared with traditional MRFs

the approach improves dense matching accuracy by a combination of adaptive disparity range predictions

adaptive matching windows

and adaptive weight coefficients. Furthermore

aMRF reduces the disparity search range while accurately preserving disparity discontinuities. Real rover images from the Mars E xploration Rover mission and Chang’E lunar orbiter images experiments demonstrate the effectiveness of the proposed method.