Mean

Calculates the arithmetic mean value of all pixels inside the reference region with no rejection of extreme values. This is the preferred method if the region contains only well-behaved statistical noise.

Maximum

Creates an image containing the maximum value of all images at each point.

Mean - Alpha Clipped

Calculates the arithmetic mean after discarding a specified number of the highest and values from the pixels in the region. For example, if the number of pixels is n = 2500, High Alpha = 20, and Low Alpha = 40, then the 20 highest and 40 lowest values are rejected and the mean value is based on a sample of 2440 pixel values. You can also set an alpha value to 0 so that no pixels are discarded at that limit.

Mean - Contra Harmonic

Computes the contra-harmonic mean value at each point. This method decreases the effect of distant pixels in the tail of the Gaussian distribution by including them in a statistically more significant result than would be obtained by rejecting them from the sample. The resulting value is based upon a nonlinear sum over all images using pixel value x and exponent, e, as

  x ^ (e+1) / x ^ (e).

A non-zero exponent biases the mean toward higher or lower values. An exponent e > 0 selectively emphasizes values higher than the mean, whereas e < 0 emphasizes values lower than the mean.

For example, a value e=2 does a good job when 10% of the images have negative outlying points.

Mean - Geometric

Computes the geometric mean value at each point. Use this to de-weight pixels more distant from the mean. This decreases the effect of deviant pixels while including them in the resulting value rather than rejecting them. Using n images, this method computes the product of all pixel values raised to the 1/n power.

Mean - Max Clipped

Creates an image with the highest value discarded from the average at each point. For a sample of n images, then, each resulting pixel value is the mean value of 19 values. Use this method when there is a good likelihood of a single outlying point, such as a cosmic ray or radioactive decay detection at any point.

Mean - Min/Max Clipped

Creates an image containing the mean value at each point after rejection of the minimum and maximum values. At each point, the minimum and maximum values are not included in computing the mean value. This is an excellent way to remove noise by discarding only 2 of the total number of values being averaged. In this method, even the statistically insignificant deviations, or "true noise" at the extremes are rejected from the mean value. This method works well for bad pixel rejection using a large number of images in which it is likely that a dark or bright non-noise pixel is likely to be found at most locations. If the number of images is small, e.g., < 5, do not use this method.

Mean - Modified Trimmed

Computes the Modified Trimmed Mean, which is a hybrid of the sigma clipped mean and median methods. See description of Mean - Sigma Clipped.

Mean - Rank Clipped

Computes the mean value including only the values inside specified rank limits. For example, usingUpper Rank = 2.0 and Lower Rank = 5.0 rejects pixel values higher than the 98th percentile and lower than the 5th percentile of the sample at each location.

Mean - Sigma Clipped

Computes the arithmetic mean after iteratively rejecting extreme values more than sigma's from the mean value. The rejection is 2-tailed, and the upper and lower sigma multipliers are specified independently. For example, setting High Sigma = 2.5 and Low Sigma = 5.0, then the statistic is computed using values within 2.5 sigma's above the mean and 5.0 sigma's below the mean.

Mean - Yp Weighted

Computes the power-weighted mean value at each point. This method decreases the effect of distant pixels in the tail of the Gaussian distribution by including them in a statistically more significant result than would be obtained by rejecting them from the sample. The historical name "Yp Mean" derives from raising "value x to the y power".

A non-zero exponent biases the mean toward higher or lower values. An exponent Yp = 0 gives a normal unweighted mean. A power Yp > 0 selectively emphasizes pixels higher than the mean, whereas Yp < 0 emphasizes pixels lower than the mean.

Median

Creates an image containing the median values of all images at each point. This method has good ability to reject extreme values. For a given number of input images, the noise in the resulting image is not as low as that which can result from Mean combining methods.

Midpoint

Computes the average of the highest and lowest values in the reference region.

Minimum

Creates an image containing the minimum value of all images at each point. This is a good way to bright artifacts that affect a large fraction of the images at the same coordinates. For example, if 5 of 8 images have large bright areas or defects, you can remove them using this method. The advantage over simply throwing away the 5 images is that there can still be some filtering done to areas that do not show the bright defects.

Rank

The rank statistic selects the pixel with the desired percentile value from the population of pixels at each location. For example, the median method selects the pixels with 50th percentile rank and the minimum method selects pixels with 0-th percentile rank. The Rank Statistic method can select any of these by setting the percentile to value in the range of 0 to 100.