Ad
Ad
Ad
Pages: [1]   Bottom of Page
Print
Author Topic: BasICColor IMProve  (Read 4319 times)
Rhossydd
Sr. Member
****
Offline Offline

Posts: 1965


WWW
« on: October 30, 2011, 10:06:12 AM »
ReplyReply

Anyone else trying this out ? or know how much it will cost ?
Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 9188



WWW
« Reply #1 on: October 30, 2011, 11:58:29 AM »
ReplyReply

Downloaded but haven’t tried it yet. Seems VERY expensive, I thought I saw it was 500 euro on the non English site.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
Rhossydd
Sr. Member
****
Offline Offline

Posts: 1965


WWW
« Reply #2 on: October 30, 2011, 12:48:26 PM »
ReplyReply

Seems VERY expensive, I thought I saw it was 500 euro on the non English site.
Wow, that's serious. I don't think it could offer enough benefit to be worth investing in at that price.

I'll be interested to hear how you think it performs and what benefits it offers.
Logged
smilem
Full Member
***
Offline Offline

Posts: 216


« Reply #3 on: November 01, 2011, 08:33:45 PM »
ReplyReply

Just tried it on i1profiler ICC profile made from my custom target. I had gray scale problems with it.
It does not seem to improve anything, quite the opposite I would say.
Curves seem better, I wonder if i1profiler targets work only with i1profiler. Anyone else tried it yet on your measurements.

before measurement optimization      after measurement optimization

dE Report

Number of Samples: 988         Number of Samples: 988

Delta-E Formula dE2000         Delta-E Formula dE2000

Overall - (988 colors)         Overall - (988 colors)
----------------------------------------------------------------------------------------------------
Average dE:   1.55            Average dE:   5.98
    Max dE:   5.79            Max dE:  14.39
    Min dE:   0.06            Min dE:   1.62
 StdDev dE:   1.05             StdDev dE:   2.37

Best 90% - (888 colors)         Best 90% - (888 colors)
----------------------------------------------------------------------------------------------------
Average dE:   1.28            Average dE:   5.38
    Max dE:   2.90            Max dE:   9.48
    Min dE:   0.06            Min dE:   1.62
 StdDev dE:   0.70            StdDev dE:   1.57

Worst 10% - (100 colors)         Worst 10% - (100 colors)
----------------------------------------------------------------------------------------------------
Average dE:   3.87            Average dE:  11.32
    Max dE:   5.79            Max dE:  14.39
    Min dE:   2.91            Min dE:   9.49
 StdDev dE:   0.76            StdDev dE:   1.29
« Last Edit: November 01, 2011, 08:37:06 PM by smilem » Logged
smilem
Full Member
***
Offline Offline

Posts: 216


« Reply #4 on: November 02, 2011, 07:47:22 AM »
ReplyReply

There seems to be a bug in their software that changes white point while converting spectral data to XYZ LAB,
The profile curves seem identical except the white point is yellow compared to normal D65.

I set the correct t whitepoint D65 while building the profiles but it made no difference, the profile still has wrong whitepoint.

The same measurement data loaded into basiccolor print builds profile with correct whitepoint and very different curves compared to i1profiler. Neutral rendering seems to be the same like i1profiler. This is true for perceptual and rel colorimetric. So the moral seems to be that basiccolor and i1profiler are not compatible, is it done on purpose or is a bug I do not know.

Here is report for basiccolor print profile made from corrected measurement data
Number of Samples: 988

Delta-E Formula dE2000

Overall - (988 colors)
--------------------------------------------------
Average dE:   1.64
    Max dE:   7.80
    Min dE:   0.07
 StdDev dE:   1.51

Best 90% - (888 colors)
--------------------------------------------------
Average dE:   1.25
    Max dE:   3.88
    Min dE:   0.07
 StdDev dE:   0.93

Worst 10% - (100 colors)
--------------------------------------------------
Average dE:   5.16
    Max dE:   7.80
    Min dE:   3.90
 StdDev dE:   0.96

--------------------------------------------------
« Last Edit: November 02, 2011, 08:16:44 AM by smilem » Logged
smilem
Full Member
***
Offline Offline

Posts: 216


« Reply #5 on: November 06, 2011, 01:54:41 AM »
ReplyReply

Come on nobody else tested this software?  Huh
Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 9188



WWW
« Reply #6 on: November 06, 2011, 10:58:22 AM »
ReplyReply

Haven't had time to test it yet. But on the Colorsync list, one person had issues and attributed it to 'bugs'.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
smilem
Full Member
***
Offline Offline

Posts: 216


« Reply #7 on: November 06, 2011, 07:10:01 PM »
ReplyReply

Haven't had time to test it yet. But on the Colorsync list, one person had issues and attributed it to 'bugs'.

I was able to build a profile using PM5, the profile was better then simply using PM5.
It seems there is a bug that i1profile builds profile with delta E 6, and PM5 can build with delta E 1.5 Huh
I don't expect i1profiler to be the best software, I mean always, but this is strange.
Logged
Ethan_Hansen
Full Member
***
Offline Offline

Posts: 114


WWW
« Reply #8 on: November 10, 2011, 01:35:39 AM »
ReplyReply

Y'all got me curious. The whitepoint problem noted by smilem happens because most of the tools convert the file data from spectral to LAB. Any such conversion requires an illuminant. This can be a standard whitepoint, e.g. D50, or an arbitrary one. The Lighting tab in i1Profiler allows choosing either a canned illuminant or a measured curve. The Viewing Light Source in PMP5 fulfilled the same function. As far as I can tell, BasICColor IMProve converts to LAB with a forced D50 illuminant.

Much of what IMProve does mirrors the optimization our own profiling code performs. Pf course, there is a fancy GUI and many more bells-and-whistles than our bare-bones number crunching algorithms. Overall, my initial impression was that IMProve needs improvement primarily in workflow. The main button is labled "Automatic...", replete with wizard's wand and dancing stars. Clicking the magical, mystical Automatic button averages any redundant values in the data file, replacing them with a single point (generally a good thing), "corrects" the data "detects faulty measurements or 'illogical' measured values and replaces these with expected values" (not so sure about this one), and "smooths" the results (unsure about this one too). Spectral data are converted to LAB at the correction step with, as far as I can tell, no way to specify an illuminant other than D50.

The redundancy and averaging function works perfectly well. If the profiling code you use does not perform internal averaging, this is beneficial. Our targets include sets of redundant patches so we can detect printing and measurement errors, and across-page output sensitivity, and as a way to obtain averaged data on such critical parameters as paper white. ProfileMaker did not do much with redundant data. I do not know how i1Profiler behaves here. IMProve's averaging is a perfectly acceptable step to perform.

I get the feeling that the correction function must assume the data are from a BasICColor target. I took one of our standard targets that we used for our in-house calibration. This target was measured four times on each of on six Spectroscans. Looking at the data from one Spectroscan, the maximum difference between any of the measurement sets was 0.57 dE-2000, with an average of less that 0.1 dE-2000. IMProve insisted that there were "measurement failures" on both each of the four individual measurement sets and the averaged file of over 4 dE (assuming dE LAB, since this is what is specified in other reports). I then took a chart that our code flagged as having actual measurement errors on an iCColor. (we have redundant patches to check chart alignment and other problems; this reading was of a 3-page target with page 1 inserted off-angle enough to corrupt the bottom few rows). Sure enough, IMProve highlighted a dozen or so patches as errors. Unfortunately there was no correlation between the marked patches and ones that had actual measurement errors.

A excellent feature of IMProve is that it allows comparing results from each step of a workflow, with the ability to highlight only those patches that were altered above a certain threshold. I did an admittedly brief evaluation of the smoothing function. The documentation indicates that smoothing is intended to compensate for measurement artifacts caused by the media rather than the printer itself. In the cases I looked at, IMProve's smoothing insisted the slightly off-neutral shadows required adjustment. No, that was how the printer really responded. Again, I don't know if this behavior was the result of throwing measurements from targets other than BasICColor's own at IMProve or if BasICColor Print works best when fed data without odd kinks. Our own code requires spectral data. Sending smoothed data into i1Profiler made for profiles that, after round trip measurements, were less accurate than from the raw data.

IMProve also sports a UV brightener detection tool. It has a slider ranging from 0% to 100% correction. According to the documentation, a 100% correction will give results equivalent to using a UV filter. I could not find a paper or RGB measurement set that IMProve did not report as having UV brighteners. These included OBA-free fine art papers, even ones measured with UV filtered instruments and the UV-cut measurements from an i1 iSis. The "corrections" for these data sets was low - 0.1 to 0.3 dE for paper white. Scaling the brightener compensation down to 75% or so gave good-looking results. Perhaps even better than our own algorithms.

UV brightener compensation has definite use for i1Profiler. If you use an iSis with i1Profiler, yes, the OBC module works. In a thoroughly impractical way if you have any number of profiles to build, but it is there. Our experience is that for papers not swimming with brighteners, the OBC module is, at best, only marginally superior to software OBA compensation. For those relying on an i1Pro, i1Profiler leaves you stuck. With a standard i1Pro, there is no compensation for brighteners. The UV-cut model happily reports data from wavelengths shorter than 400nm, although the UV filter effectively chops these to nothing. Where these values come from, only X-Rite knows, but they certainly are not actual measurements. BasICColor IMProve offers a solution, but the $675 price tag is steep if all you need is an OBA compensation algorithm.

IMProve has several other features I only explored briefly. White and black correction tools can be used to pre-edit the profile data to alter K-Gen curves or paper white points. Could be useful. An ICC transformation tool allows simulating the combined effects of up to three standard, devicelink, or saveink (a BasICColor exclusive) profiles using the measurement data. Again, I can see some use for this.

Finally, I must be missing something when it comes to the Rescale tool. Rescale offers a way to interpolate and/or extrapolate a limited set of measurement data into a larger target. The example given in the documentation is for a printing process lacking the uniformity to print a full ECI2002 chart (speculation as to what the final copy from such a process look like is left to the reader). With the magic of Rescale, you can print a small target or two, scale it up to a larger one, and profile away. I experimented with a RGB set. I extracted a 729 patch data set (points spaced 32 RGB units apart) and ressed it up to 4160 patches (16 RGB point spacing + 64 grayscale steps). i1Profiler took the data and made a mess of a profile. Perhaps BasICColor Print works better with closer spaced input. I just don't know. The process strikes me as the profiling equivalent of photographers making huge prints from point-and-shoot cameras. Gee, you mean had I only discovered Bicubic Smoother, we sure could have saved some money on IQ180 backs?

I'm left with decidedly a mixed impression of IMProve. The redundancy tool is excellent, if you are measuring targets with repeated patches. Properly tamed, the UV Brightener compensation tool worked well. Perhaps the correction and smoothing tools work with BasICColor targets or software. They do presuppose a D50 illuminant for the profile. The other tools ranged from useful for specialized cases (white/black, ICC transform) to WTF?!? (Rescale). The ability to compare results at each step of the process is excellent, and going back through the history stack to try something else is easy. For an automated workflow, IMProve would benefit from the ability to chain together an arbitrary flow of steps (macros, anyone?). Overall, I see the most useful tool being the OBA compensation. Package that as a standalone module, perhaps with the redundancy module as a front end, price it accordingly, and it would be attractive for i1Profiler users.
Logged

smilem
Full Member
***
Offline Offline

Posts: 216


« Reply #9 on: November 10, 2011, 01:30:14 PM »
ReplyReply

Quote
The UV-cut model happily reports data from wavelengths shorter than 400nm, although the UV filter effectively chops these to nothing. Where these values come from, only X-Rite knows, but they certainly are not actual measurements.

i1Pro has UV filter on the light source not the actual sensor. So the readings are from spectrum that is not agitated by UV light.
Logged
Ethan_Hansen
Full Member
***
Offline Offline

Posts: 114


WWW
« Reply #10 on: November 10, 2011, 06:40:33 PM »
ReplyReply

i1Pro has UV filter on the light source not the actual sensor. So the readings are from spectrum that is not agitated by UV light.

Not quite. As you mention, the i1 Pro has a yellow outer filter that largely eliminates all light output at wavelengths shorter than 400nm. The lamp is a gas-filled tungsten, with an output spectrum roughly approximating Illuminant A (~2850K). As you can see from the plot below, these lamps have very little output in the long-wave UV A region below 400nm. The combination of lamp output characteristic and filter makes the UV contribution to the light source nearly non-existent.

If you have a dead UV-cut i1Pro handy to operate on, open it up and you will find the filter has two parts. The outer ring is the above-mentioned yellow filter for the light source. The inner ring contains an additional UV-cutoff filter. This filter effectively blocks light with wavelengths under 400nm. So the instrument shines little or no UV light on the target, and any UV emissions that may occur are filtered out.


Now, open a measurement file made using a UV-cut i1Pro. The old-fashioned CGATS format from PMP or MeasureTool is easiest to read. Look at a paper white measurement. Lo and behold, there are large numbers in the 380 and 390nm columns! Even more curiously, these values are often larger than what you get with a standard, UV-included i1 Pro. X-Rite must be extrapolating (read manufacturing out of whole cloth) data in the shortest wavelengths. Because the lamp emits no UV light and the measurement filter chops out what little remains, all that is left is noise. X-Rite's algorithm goes nuts with the data. A better approach would be to truncate the readings to remove all data below 400nm for UV-cut instruments.

This behavior is further justification for why UV-cut measurements really are a poor option, particularly with X-Rite instrumentation. Filtering UV is, from a color science perspective, difficult to justify when other options are available. Software compensation is almost always superior, while the iSis OBC module provides an even better approach.

Here's a real world example. The plots below compare paper white readings on a stock with the barest hint of optical brighteners. PMP5's software algorithms did not detect any brighteners, while our code dials in a compensation of 10% of maximum. The plot on the left is from a UV-cut i1Pro, the right graphic from a standard instrument. In each plot, the leftmost two bars are 380 and 390 nm. You can see the no-cut measurements show a minor peak at 440nm that is smoothed out in the UV-cut data. In fact, all UV-cut measurements from 420 to 500nm, have lower values than the unfiltered measurements. Given that these were made by i1 Pro's - not the most accurate instruments by a long shot - this may simply be differences in how the two devices read.

Note, however, the large increase in the value of the 390nm point (second bar from left) in the UV-Cut measurements and the even larger bump at 380nm. That is not reality. It's extrapolation gone amok.


Logged

smilem
Full Member
***
Offline Offline

Posts: 216


« Reply #11 on: November 11, 2011, 08:57:40 AM »
ReplyReply

Quote
The inner ring contains an additional UV-cutoff filter.

I did some experiments with i1pro UV cut for fluorescence testing, and I had to hand pick out of 10 or so the right external UV-cut filter to cut up to 400nm. I have not X-rite, but gretagmacbeth device maybe x-rite changed something.

But I say my device can read the little UV range in emission mode. Are you saying there is a filter for incoming light into the device to cut UV even in emission mode??? Then how come I needed the external filter?
Logged
Pages: [1]   Top of Page
Print
Jump to:  

Ad
Ad
Ad