To suppress speckle noise efficiently and achieve less time consuming in laser active imaging, a denoising method based on spectrum-domain constraint in signal subspace is proposed. Homomorphic transform is performed to convert the multiplicative noise to additive noise. The noisy image is decomposed using singular value decomposing to estimate the dimension of signal subspace. Based on the dimension, the covariance matrix of the noisy image is decomposed using eigenvalue decomposing to estimate the covariance of speckle noise with the eigenvector in noise subspace. After that the denoising estimating matrix is computed with the eigenvector in signal subspace and is convolved with the noisy image. Inverse homomorphic transform is carried out and the denoised image is obtained. In order to eliminate the effect of backscatter and background radiation, a laser active imaging system based on range gating ICCD is constructed. The experiment is performed using images with speckle noise in different levels and the results show that compared with classical Lee, Frost and Kuan filters, the proposed method has advanced denoising performance and the computing time is in millisecond level, which meets the requirement of real-time operation for laser active imaging.