Noise detection-based median filters have been widely adopted to reduce salt and pepper noise in images. However, since noise pixel is not detected accurately, it is likely to blur the fringe of image under the case of high noise density. An algorithm of salt and pepper noise filter is proposed that is based on GA-BP algorithm noise detector to remove the salt and pepper noise in images. The algorithm firstly detects noise pixel’s location in image by using optimized GA-based BP network. Then, it introduce edge-preserving function and PRP algorithm to solve the objective function of extreme value and realize the image denoising. Compared with the traditional algorithms, experimental results show that the proposed algorithm has an evident improvement, and good characters of generalization, robust and self-adaptive.