Infrared images typically have the problems of target fuzzy edges and details. The variable infrared image enhancement algorithm of bimodal Gaussian function specification is introduced. Firstly, by converting the image to the gradient domain, image gradient histogram can be obtained. Then, by constructing a bimodal Gaussian function, the distribution of the gradient histogram can be restricted. Finally, by using the variational method to reconstruct the image from the gradient field, the contrast and target edges and details of the image are improved. The cutting of the amplitude of noise in the transform of gradient field prevents the amplification of noise. Based on actual experimental results, both in visual effects and quantitative indicators of the assessed value of the image information entropy, the algorithm is significantly better than the histogram equalization and specification. Therefore, it gives a good visual effect for the infrared image.