ISO TS 23564:2020 pdf free.Image technology colour management一Evaluating colour transform accuracy in ICC profiles.
4 Profile evaluation tests
4.1 General
Two types of profile evaluation tests may be performed on the media-relative colorimetric rendering intent (the BToA1Tag and the AToB1Tag) of an ICC profile. These tests apply the transform contained in the profile’s BT0A1Tag and/or AToB1Tag and compare the results with an appropriate reference.
4.2 Round trip tests
Round trip tests determine the accuracy with which a given rendering intent within a profile is inverted, such that when a PCS value is converted to the device encoding and back to the PCS the difference is minimized. When performing a round trip test, it is essential that the final test is performed on coordinates within the gamut of the encoding, and an initial transform from the device data encoding may be required to achieve this.
4.3 Device model tests
Device model tests of a profile determine the accuracy of the forward model with which the profile predicts the colorimetry of a given device encoding value, or the accuracy with which the profile predicts the device encoding value that will produce a given colorimetry.
See Annex A for additional information on forward model accuracy.
5 Round trip accuracy
5.1 General
A round trip test of a profile is performed by converting a set of PCS values to the device encoding, using the profile’s I3ToAlTag, and then back to PCS values using the profile’s AToB1Tag.
5.2 Test image
The set of PCS values for the round trip test shall be obtained as follows. A test image such as that defined in ISO 12642-2 is used to provide a sample of the device encoding. This image is converted to PCSLAB using the AToB1Tag of the profile, back to device space using the BT0A1Tag, and finally back to PCSLAB using the AToB1Tag. This procedure ensures that only in-gamut colours are used in the round trip accuracy test.
Where possible, a test image should contain different values from those used to derive the AT0B1Tag tables and from those used as the input table in the BToA1Tag transform.
NOTE The differences in PCSLAB values between the first and second set of PCSLAB values obtained in this way are often referred to the “first round trip” accuracy.
In addition to the test chart defined in Iso 12642, particular applications may require additional test colours representative of the sub-set of the colour gamut of interest to that application. Examples of such sub-sets include additional neutrals and near-neutrals to better evaluate the grey balance of a transform; and colours corresponding to skin colours for images where such colours predominate. In such cases, a user- or vendor-dependent test set may be added. The evaluation report shall state how many test colours and which test image was used in the evaluation.
5.3 Roundtriptest
The final PCSLAB values obtained in Li are used as the initial values in the round trip test. They are converted to device space using the BToA1Tag, and then back to PCSLAB using the AT0B1Tag.
The differences between the initial PCSLAB values and the PCSLAB values after the round trip shall be expressed as median, maximum and 95th percentile values in E00, where the PCSLAB values are scaled to ICC-Absolute colorimetry according to ISO 15076-1. Round-trip colour differences in tE00 typically are < 1 median and < 2 maximum. Preview tags are not used for testing round-tripping errors. NOTE 1 Round trip results are likely to be affected by the device characteristics and the colour management module (CMM) used, as well as the profile. In a comprehensive test, a profile that is to be used in open systems is evaluated using several CMMs, preferably those that will be used in practice.
NOTE 2 The round trip error does not predict the accuracy of the forward or inverse model, but is simply a measure of the inversion accuracy of the profile. It can be used to estimate the inverse model error as described in Clause 6.
NOTE 3 The utility of the round trip error is directly related to the accuracy of the forward model fit. The round trip error is less meaningful if the forward model is poor, for example if there is noise in the training data used to generate it.ISO TS 23564 pdf download.