Intensity Transform Similar to DS9 'zscale' ?
- Thread starter cgibson
- Start date
FWIW, this is a summary of the IRAF 'zscale' transform, which is embedded in SAO's DS9 tool:
The zscale algorithm is designed to display the image values near the median image value without the time consuming process of computing a full image histogram. This is particularly useful for astronomical images which generally have a very peaked histogram corresponding to the background sky in direct imaging or the continuum in a two dimensional spectrum.
The sample of pixels, specified by values greater than zero in the sample mask zmask or by an image section, is selected up to a maximum of nsample pixels. If a bad pixel mask is specified by the bpmask parameter then any pixels with mask values which are greater than zero are not counted in the sample. Only the first pixels up to the limit are selected where the order is by line beginning from the first line. If no mask is specified then a grid of pixels with even spacing along lines and columns that make up a number less than or equal to the maximum sample size is used.
If a contrast of zero is specified (or the zrange flag is used and the image does not have a valid minimum/maximum value) then the minimum and maximum of the sample is used for the intensity mapping range.
If the contrast is not zero the sample pixels are ranked in brightness to form the function I(i) where i is the rank of the pixel and I is its value. Generally the midpoint of this function (the median) is very near the peak of the image histogram and there is a well defined slope about the midpoint which is related to the width of the histogram. At the ends of the I(i) function there are a few very bright and dark pixels due to objects and defects in the field. To determine the slope a linear function is fit with iterative rejection;
I(i) = intercept + slope * (i - midpoint)
If more than half of the points are rejected then there is no well defined slope and the full range of the sample defines z1 and z2. Otherwise the endpoints of the linear function are used (provided they are within the original range of the sample):
z1 = I(midpoint) + (slope / contrast) * (1 - midpoint)
z2 = I(midpoint) + (slope / contrast) * (npoints - midpoint)
As can be seen, the parameter contrast may be used to adjust the contrast produced by this algorithm.
The zscale algorithm is designed to display the image values near the median image value without the time consuming process of computing a full image histogram. This is particularly useful for astronomical images which generally have a very peaked histogram corresponding to the background sky in direct imaging or the continuum in a two dimensional spectrum.
The sample of pixels, specified by values greater than zero in the sample mask zmask or by an image section, is selected up to a maximum of nsample pixels. If a bad pixel mask is specified by the bpmask parameter then any pixels with mask values which are greater than zero are not counted in the sample. Only the first pixels up to the limit are selected where the order is by line beginning from the first line. If no mask is specified then a grid of pixels with even spacing along lines and columns that make up a number less than or equal to the maximum sample size is used.
If a contrast of zero is specified (or the zrange flag is used and the image does not have a valid minimum/maximum value) then the minimum and maximum of the sample is used for the intensity mapping range.
If the contrast is not zero the sample pixels are ranked in brightness to form the function I(i) where i is the rank of the pixel and I is its value. Generally the midpoint of this function (the median) is very near the peak of the image histogram and there is a well defined slope about the midpoint which is related to the width of the histogram. At the ends of the I(i) function there are a few very bright and dark pixels due to objects and defects in the field. To determine the slope a linear function is fit with iterative rejection;
I(i) = intercept + slope * (i - midpoint)
If more than half of the points are rejected then there is no well defined slope and the full range of the sample defines z1 and z2. Otherwise the endpoints of the linear function are used (provided they are within the original range of the sample):
z1 = I(midpoint) + (slope / contrast) * (1 - midpoint)
z2 = I(midpoint) + (slope / contrast) * (npoints - midpoint)
As can be seen, the parameter contrast may be used to adjust the contrast produced by this algorithm.
This would be easy to implement in PixInsight, but I don't see the benefits. Our automatic display function algorithm (AutoSTF) can provide a much more useful and controllable visualization of linear images. Can you elaborate on the reasons why you might prefer the above method?
fredvanner
Well-known member
I think the "zscale" function is the IRAF equivalent of the STF - a quick-look, low-cost visualisation tool. It was implemented more than 10 years ago, when CPU load was more of an issue, and has various methods of sub-sampling the image to reduce load. As Juan notes above, the STF will almost always be a better process to meet the same objective.