Return to Previous Section

C. Medianing of Bias Frames


At a given temperature, one can obtain a much better realization of the 2-D pattern in the bias by combining a series of individual bias frames. The best method for doing so is to take an odd number, N, of bias frames, and combine them with an algorithm that examines the N values for each pixel location, and uses the median of those N values for that pixel location on the output image. This is typically referred to as "taking the median of the images" or "medianing the images". Medianing is better at removing stray hot pixels and such than is averaging. As the number of frames increases, the resulting noise in the median will decrease, in principle as the square-root of N, where N is the number of frames. In practice, the reduction will be more modest than this, but still substantial. Figure 5 shows a set of two bias frames. The first is an individual bias frame, taken at T ~ -20 degrees C. The second is the median of 13 individual biases, all taken at T ~ -20 degrees C. They are displayed at the same contrast.

  • Figure 5: A single bias frame (top panel) compared to a median of 13 bias frames (bottom panel). All frames obtained at T ~ -20 degrees C.

    To provide a more quantitative demonstration of the impact of medianing a set of biases, I defined a 300 by 300 pixel region in the center of the frame (columns 601-900, rows 351-650), and evaluated the mean and dispersion of the pixel values in the single frame, and the medianed frame. One should not use the entire images for such a test, as the large-scale 2-D structure in the bias will dominate the dispersion if one uses the entire image. To evaluate the pixel-to-pixel noise one must use a small enough region to avoid any substantial gradients in the bias level, but a large enough region to get good statistical sampling. The results of this test are shown in Table 1. Note that the result of medianing in this case is to produce a bias frame with a noise level a factor of ~2.4 lower than is obtained for a single image. This is a smaller factor than the ideal case (for N=13, the square-root of N is ~3.6), but still a substantial improvement compared to the single frame.

    Table 1 - Comparison of Single and Medianed Bias Frames
    Frame Region Mean Dispersion
    Single Frame 107.59 4.25
    Medianed Frame 108.32 1.75


    Procede to Next Section

    Return to Main Bias Page

    Return to Outline


    Updated: 2007 September 3 [pbe]