Narcissus effect was an important factor that influence the imaging quality of the cooled infrared detector. Although it can be removed by non-uniform correction techniques, narcissus effect will reappear once the operating conditions have changed. Through analyzing the manifestation of narcissus effect, a correction algorithm based on adaptive spatial-temporal filtering was proposed. First, spatial domain estimation of Narcissus effect was captured by wavelet transform. Then, time domain estimation of Narcissus effect was gained by adaptive time domain low-pass filtering. Finally, the correction was applied by subtracting the estimated spatial-temporal noise from the original image. Experimental results on simulated and actual infrared image sequences have verified the effectiveness of the proposed algorithm.