Abstract: Given that the low resolution and poor texture information of passive millimeter wave images, a pre-selection based image de-noising algorithm(PBNL) was proposed in this paper. The local regional characteristics of the image were obtained by using the singular value decomposition(SVD) of image gradient information, then the image was divided into different categories, and relevant algorithms were adopted. Experimental results show that, in contrast to the non local means(NL-Means) algorithm, the computational time complexity of the method proposed in this paper is significantly reduced and peak signal-to-noise ratio(PSNR) is superior to current state-of-art denoising algorithms, such as BM3D, anisotropic denoising algorithm, and it can get a better recognition results visually.