In order to select the appropriate threshold in wavelet de-noising method,a new method for image de-noising based on threshold self-adjustment was proposed in the paper.The algorithm combined wavelet with the nonlinear hyperbolic tangential function in BP neural network based on the conventional wavelet threshold method.Firstly,the wavelet of noisy image was decomposed by binary wavelet transform;Secondly,the partial wavelet coefficients were reconstructed,and the gradient descent method in BP neural network was used to optimize the reconstructed wavelet coefficients to search the optimal threshold;Finally,the new reconstructed wavelet coefficient through threshold was gotten,it was the estimated value of the original image.Experimental results show the better reconstructed image visual effect and demonstrate that the signal-noise-ratio(SNR) and the peak signal-noise-ratio(PSNR) of this algorithm have been improved 1-2 dB in comparison with common wavelet threshold method.