Image segmentation is usually needed to extract the features in order to recognize images of terahertz inline digital holography reconstruction. The images of terahertz inline digital holography reconstruction contain various degree of noise, which affects the quality of segmentation. In this paper, the composite segmentation method based on the clipping, mirror expansion, filtering and histogram polynomial fitting is adopted for the characteristics of terahertz holographic reconstructed image. By using this method, the 2.52 THz images of terahertz inline digital holography reconstruction of gasket and gear are processed and compared step by step with the basic global threshold method and the Otsu method, and objectively evaluated with Mathews correlation coefficient. Experimental results show that the composite method can get better segmentation results.