Nikolay Pokhilchenko wrote:
What do You think about invprofcheck result while profiling? My own experience tells me that in general case the invprofcheck errors must be the same size as profcheck errors. If the invprofcheck errors are larger than profcheck errors, the profile is likely overfitted and colprof -r value must be icreased. And vice versa. The equivalence of
You can't tell how well a profile fits a device without taking an independent set of measurements as a reference. The colprof report will tell you how well the profile fits the data, but you don't know how well the data represents the device.
inv- and profcheck results asserts that profile fitting are well-balanced and both B2A and A2B tables are quite exact.
invprofcheck only checks the A2B table against the B2A table, and doesn't make any reference to the original test data, so it only gives an indication of the loss of precision in the way the B2A table is formed. The numbers it produces are rather sensitive to how it determines that colors are out of gamut though. If color quality is critical, don't use the B2A table, it is inherently a poor way of transforming color since it grids out the whole PCS, wasting lots of table entries or demanding a very high grid resolution which consumes lots of processing power to invert mostly out of gamut grid values. Instead create a device link using collink -G. A device links CLUT maps from in gamut to within gamut, making much better use of your profile size and CPU cycles. It's easy to create a profile aimed at being used with collink -G, with a high res. A2B table and low res. B2A table by using the coloprof -b option. [This assumes of course that a device link is usable within your workflow.] Graeme Gill.