Selective Temporal Averaging

This plugin is designed to smooth out non-moving areas of a video clip (figure 10.60).

Figure 10.60: STA control window
Image staveraging

Denoise is generally done on a spatial basis, mediating the values of a group of adjacent pixels to achieve greater uniformity. The effectiveness of Denoise can be increased by also introducing a time average between a group of successive frames. The union of these two phases is the basis of the plugin. In fact the smoothing is performed by averaging the color component for each pixel across a number of frames. The smoothed value is used if both the standard deviation and the difference between the current component value and the average component value are below a threshold. The standard deviation is a mathematical index used to estimate the variance of a group of pixels: at high values corresponds more variation of the pixels and therefore more noise.

The Selective Temporal Averaging plugin plays on the homogenization of the values of a group of pixels in a group of frames, based on a threshold below which the original values are left and above which the average is performed and then the noise reduction.

The average and standard deviation are calculated for each of the components of the video. The type of components averaged is determined by the color model of the entire project. The average and standard deviation of the frames can be examined by selecting the specific radio button in the plugin options window. The region over which the frames are averaged is determined by either a fixed offset or a restart marker system. In a restart marker system, certain keyframes are marked as beginning of sections. Then for each section, the frames surrounding the current frame are used as the frames to average over, except when approaching the beginning and end of a section, where the averaging is performed over the N beginning or ending frames respectively.

An example of common usage is to select the number of frames you wish to average.

  1. Enter a reasonable number of frames to average (for example, 10).
  2. Select the Selective Temporal Averaging method and enter 1 and 10 for all the Av. Thres. and S.D. Thres. respectively. This basically causes all pixels to use the average value.
  3. Turn the mask for the first component on. This should make the whole frame have a solid color of that specific component.
  4. Slowly reduce the S.D. Thres. value. As you do so, you will notice that the regions vastly different from the average will have a flipped mask state. Continue to reduce the threshold until you reach the point at which non-moving regions of the video have a flipped masked state. This value is known as the noise-floor and is the level of natural noise generated by the CCD in the camera.
  5. Repeat the same procedure for the Av. Thres.
  6. Turn off the mask.
  7. Repeat this for all channels.

The CINELERRA-GG Community, 2021