Abstract: In this paper, an improved adaptive median filter algorithm was proposed to solve the problem of high complexity of adaptive median algorithm. First, the filter window size was determined according to the distance between the valid pixels and the center pixels in the proposed algorithm, which can avoid the waste of pixels repeated sort in window expand process. Second, the median value pixels were taken from valid pixels within the window by the improved algorithm, which effectively weakened the interference of noise point. The quality of image was improved compared with the original algorithm. This proposed algorithm reduced the complexity of algorithm and improved the effectiveness of noise reduction. The average PSNR value was increased 10dB. Finally, the algorithm was applied to noise reduction with depth image captured from Kinect.