Aiming at the shortcoming of extremum median filter in denoising the salt and pepper noise, an improved image filtering algorithm is proposed. The extremum method is used to detect the noise pixel firstly, and then an adaptive multiwindow method is adopted to get the filtering value of pixels that is labeled as noise. The modification process is introduced for the wrongchecked pixels by comparing their gray value before and after the filtering. The proposed method is proved to be superior to the traditional median filter and some improved algorithm for the different noise image. The performance of the method is saliently improved in eliminating the noise and preserving the edge of the image when the noise ratio is high.