Terahertz time domain spectroscopy has been widely used in both spectral analysis and imaging applications. Existing terahertz time domain spectroscopy imaging systems usually suffered the low scanning speed and huge data storage. To solve this problem, an efficient terahertz imaging method based on the compressed sensing theory was presented in this paper. By controlling the scanning motor to perform a nonequal interval sampling of the target, a group of under-sampled terahertz signal can be obtained. Based on the under-sampled signal, the compressed sensing algorithm is applied to reconstruct the complete terahertz image. The results show that, when the compression ratio is 0.5, the correlation coefficient between the reconstructed terahertz signal and the fully sampled THz signal can reach 99.95%. By analyzing the reconstructed terahertz image, the image areas with smooth intensity changing or low frequency component in frequency domain can be well recovered. The proposed method provides a practical means for efficient terahertz imaging applications.