Abstract: This study puts forward an image segmentation algorithm for passive terahertz images. First of all, the captured original terahertz images are denoised in the preprocessing course. Afterwards, we select the seed regions and the growing criteria by analysis and curve fitting of the gray histogram of the denoised images, and then carry out region growing. Experiment results show that the proposed algorithm is able to extract regions of interest from background effectively, which could help to detect the contrabands hidden under the clothes of the target subjects quickly and accurately, and consequently strengthen the practicability of our imaging system.