When working with mosaics, the aligned panels usually contain lots of zero pixels. The auto-STF that is generated for these is useless

I am aware of this problem. The PI core application maintains a number of data items for all images, including statistical and histogram data among others, which we call

*global data*. Global statistics are computed and maintained (that is, recomputed only when necessary) automatically by the core and shared with all tools and processes requiring them. This greatly improves efficiency of the whole platform because computing statistics is a relatively expensive operation.

The problem is that global statistics don't use pixel rejection, that is, statistical values are computed for the actual pixels of the images, including zeros and ones. The STF AutoStretch feature uses global statistics, so now you know why it does not reject zero pixels.

The solution to this problem is replacing the current fixed scheme of global objects with a more flexible cache of module-defined global objects. This will be available during the version 1.8 cycle. A related additional improvement is also planned: optional 20-bit global histograms, which are really necessary for HDR images, where the current 16-bit histograms lead to posterization of screen renditions. This will have to wait until PI 1.9 or perhaps PI 2.0.