A new target detection algorithm is presented which deals with the problem of detecting target in remote sensing image. Based on the assumption that the distribution of the homogenous land surface is a Markov random field, we analyzed that the distribution of the homogenous land surface background is Gaussian. This model leads to an efficient and effective detector for discriminating man-made objects in real remote sensing imagery. First, the original image is transformed into Gaussian space as the ideal background. And then, the residual image is obtained by subtracting the ideal background from the original images. Finally, a conventional detector is applied to the residual image to complement the further target detection. Because the targets have values deviating significantly from the distribution of the background, the background can be severely decreased during the subtraction. Therefore, the new algorithm has better performance. Some experiments of real remote sensing images proved the validity of the new algorithm.