In order to improve the spatial resolution and reduce l/f noise of real-time THz pulses （T- ray） imaging system, a new method based on wavelet de-nosing and Jasson-Van-Cittert （JVC） image restoration algorithm was presented. After making wavelet transform to a T-ray image, generalized cross validation （GCV） was used to estimate the optimal threshold and de-noising was done in the high frequency sub-bands respectively. The JVC image restoration algorithm was applied to the de-noised T-ray image for resolution improving. Experimental results show that this method can suppress l/f noise successfully and improve the resolution of T-ray image significantly.