Median filtering is a spatial-order-statistics based image denoising method,which can preserve image details well while eliminating noise. In traditional median filters,the pixels in a square window is often sorted by bubble sorting,which demands large comparison times and long processing time,and is unfit for real-time image processing. Based on image parallel processing technology,a fast median filtering algorithm based on extended window is presented. In the proposed algorithm,the traditional square window is extended transversely,so the sliding step of window roaming can be increased by pixels to achieve much higher processing speed for the whole image,while the comparison results within the process of executing the algorithm are effectively used. The result shows that the algorithm can achieve high precision and real-time ability in image denoising. The algorithm improves the image processing speed by the cost of more area of hardware chip,and it is suitable for implementation on hardware such as field programmable gate array.