A contrast enhanced image fusion algorithm is proposed to improve the visual effect of fused image from infrared and visible images at low lightness. Estimating background luminance by bilateral filter, the subband-decomposed multiscale Retinex with bilateral filtering is proposed, and original images are decomposed using proposed transformation. Coefficient matrixes in the same subband are fused by using detail enhancement strategy based on local space frequency, and fused coefficient matrixes in different subbands are fused by global variance weighted sum. The fusion result is needed for the transition from the logarithmic domain to the display domain by improved nonlinear stretching. The results of experiments demonstrate that the proposed algorithm can effectively overcome the halo phenomenon, improving definition of fused image, and the detail information is more salient.