An denoising algorithm of convolutional neural network（ CNN） based on actual scene image is proposed. By building a new database of noiseless images

then by inputting it into CNN

and combining with simulated annealing algorithm to improve training rate

finally building a model which can realize actual scence image denoising. Experimental results show that both gray images with noise

and images by camera achieve effectively smoothing effect

power signal-to-noise ratio（ PSNR） value is high and image edges and details are preserved well.