When we need to analyze video images, we sometimes capture one still image. But the video is 30 FPS (Frame per Second). But using only one static image to represent this video is unreasonable. Moreover, we must consider the impact of temporal noise. Our method is based on SNR (Signal-to-Noise Ratio) with time relevance of each pixel. Many noise reduction algorithms have many trade-offs. We need a good measurement method. Therefore this thesis proposes some noise measurement methods and provides one new measurement mode.