1. EVALUATION OF COLOUR DIFFERENCE FORMULAE BY SPANISH OBSERVERS Prof. Dr. José Valldeperas Prof. Josep M. Gibert INTEXTER- ETSEIT Universitat Politècnica de Catalunya Terrassa - Spain

2. A little of history Since 1970 the Spanish Colourfastness Committee (CES) is being involved in ECE & ISO/TC38/SC1 meetings and activities Since 1970 the Spanish Colourfastness Committee (CES) is being involved in ECE & ISO/TC38/SC1 meetings and activities 1975: WG Colour Measurement WG created into CES 1975: WG Colour Measurement WG created into CES First work: Standards UNE 40080 (colour coordinates) and UNE 40081 (colour differences, ANLab & CIELab) First work: Standards UNE 40080 (colour coordinates) and UNE 40081 (colour differences, ANLab & CIELab) From 1976, WG actively involved in every proposal made within ECE, ISO AND CEN. From 1976, WG actively involved in every proposal made within ECE, ISO AND CEN.

3. First evaluation of colour difference formulae (1995) CMC 2:1 vs. CIELAB 1976 CMC 2:1 vs. CIELAB 1976 48 pairs of real samples; 13 qualified observers 48 pairs of real samples; 13 qualified observers Visual assessment of DEv, DLv, CDv and DHv, by using Grey Scale (change in colour) Visual assessment of DEv, DLv, CDv and DHv, by using Grey Scale (change in colour) Linear regression comparison between instrumental and visual colour difference data Linear regression comparison between instrumental and visual colour difference data Result: Better performance of CMC2:1 than CIELab

4. Current ISO Colour difference standard for textiles: ISO 105 J03: 1995 FORMULA : CMC 2:1 FORMULA : CMC 2:1

6. New evaluation of colour difference formulae 2001. Mölndal (Sweden). ISO/TC38/SC1/WG7 recommended to study new formula CIEDE2000, possible future substitute of CMC2:1 2001. Mölndal (Sweden). ISO/TC38/SC1/WG7 recommended to study new formula CIEDE2000, possible future substitute of CMC2:1 Between 1992 and 2000, other new formulae appeared: CIE94, LCD, CIEDE2000, etc. Between 1992 and 2000, other new formulae appeared: CIE94, LCD, CIEDE2000, etc. A new trial was designed (last 3 + CIELAB, CMC2:1) A new trial was designed (last 3 + CIELAB, CMC2:1) Enlarged no. of samples: old 48 + new: Total 106 samples Enlarged no. of samples: old 48 + new: Total 106 samples Covering all hues, especially blue region Covering all hues, especially blue region

8. Objectives of the evaluation Main objective: To compare new CIEDE2000 with established CMC 2:1 and other formulae Main objective: To compare new CIEDE2000 with established CMC 2:1 and other formulae Other objectives: Other objectives: To test the performance in the blue region of colour space To test not only total difference, but also its 3 components (DL, DC, DH)

9. Experimental (I) A set of 106 pairs of textile samples (maximum difference DE=4) A set of 106 pairs of textile samples (maximum difference DE=4) Instrumental measurements and calculations: Instrumental measurements and calculations: Spectraflash 300 Datacolor, specular included Samples folded 4 times, background filter paper Large Area of View (22 mm diameter) X Y Z, and L*, a*, b* values: (ColorTools software: ISO 105-J01:1997, illuminant D65, standard observer 10º CIE1964) Colour differences: Colour differences: According to published formulae and data: worksheets Lotus 123; Excel, ASTM E305

10. Chroma range: >80% samples with C 80% samples with C<+-20

11. Lightness range: L=90 - 15 (pale greys to dark navies)

12. Experimental (II) Visual assessments: 8-13 observers Visual assessments: 8-13 observers Pairs of samples 30x50 mm, edge to edge Pairs of samples 30x50 mm, edge to edge Viewing cabinet: True-Vue 2 Datacolor (bottom L=77) Viewing cabinet: True-Vue 2 Datacolor (bottom L=77) At the right side of sample, GS change in color (plates 5 to 3) At the right side of sample, GS change in color (plates 5 to 3) Source simulating D65 standard CIE illuminant Source simulating D65 standard CIE illuminant Maximum contrast used : Index 3 GS Maximum contrast used : Index 3 GS Pairs showing DE(CIELAB)=>5: excluded Pairs showing DE(CIELAB)=>5: excluded

14. Table 1. Method for assigning a numerical value for the different parameters used Grey Scale rating54-543-43 DEvis01234 Difference of lightness, DLvis: Grey Scale rating33-444-55 43-43 DLvis-4-3-201234 DarkerUndistinguishableLighter Difference of chroma, DCvis: Grey Scale rating33-444-55 43-43 DCvis-4-3-201234 DullerUndistin- guishable Brighter, more saturated Difference of hue, DHvis: Grey Scale rating33-444-55 43-43 DHvis-4-3-201234 More contrast, Clockwise direction Undistin- guishable More contrast, anticlockwise direction Difference of colour, DEvis:

15. Experimental (III) Assignment of mathematical sign in DHvis: Assignment of mathematical sign in DHvis: Observer locates Reference (R) and sample (S) points in CIELab diagram Observer locates Reference (R) and sample (S) points in CIELab diagram Following the shorter way to go from R to S: Following the shorter way to go from R to S: clockwise: DHvis negative (-1... -4) clockwise: DHvis negative (-1... -4) anticlockwise: DHvis positive (+1...+4) (hue angle increases) anticlockwise: DHvis positive (+1...+4) (hue angle increases) No difficulties in assigning math sign for DLvis and DCvis No difficulties in assigning math sign for DLvis and DCvis

16. Results and discussion The arithmetic mean of visual assessments were determined for all the samples: DEvis, DLvis, DCvis DHvis These statistical values are representative of an average observer Table 1. Statistics of visual assessments ARITHMETIC MEANSTANDARD DEVIATION DLvisDCvisDHvisDEvisDLvisDCvisDHvisDEvis Max. 2,63 3,25 3,133,631,952,102,121,05 Min.-3,15-2,77-3,130,300,000,000,290,27 Mean-0,10-0,08-0,181,650,820,930,880,62 No. pairs 106 106 106106106106106106 No. observers8-13

17. Correlation colorimetric-visual DE

18. DL DL

19. Correlation colorimetric-visual: DC

20. Correlation colorimetric-visual: DH

21. Table 3. Results of overall linear regressions CORRELATION COEFFICIENTS, r VISUAL COLORIMETRICDLvisDCvisDHvisDEvis CIELAB DL0,8180 DC0,6901 DH0,7093 DE0,7051 CMC21 DL0,8257 DC0,6539 DH0,7466 DE0,7920 CIE94 DL0,8180 DC0,6551 DH0,7546 DE0,7707 CIEDE2000 DL0,8092 DC0,6597 DH0,7767 DE0,7981 DE LCD DE0,6778

22. Results and discussion Correlation between colorimetric and visual colour differences data is statistically significant, for all the formulas tested. Correlation between colorimetric and visual colour differences data is statistically significant, for all the formulas tested. The best correlation is found for DL and DE, particularly for CMC 2:1 The best correlation is found for DL and DE, particularly for CMC 2:1 DL(CIELAB) = DL(CIE94), so r=0,8180 in both cases DL(CIELAB) = DL(CIE94), so r=0,8180 in both cases CIEDE2000 gives similar results than CMC2:1, correlation does not improve CIEDE2000 gives similar results than CMC2:1, correlation does not improve DCvis is difficult to establish, then poorer agreement DCvis is difficult to establish, then poorer agreement DHvis is not so difficult to assess, but DHcolorimetric has a very narrow range, so that the slope of regression line is smaller than 1 (ideal 1, diagonal) DHvis is not so difficult to assess, but DHcolorimetric has a very narrow range, so that the slope of regression line is smaller than 1 (ideal 1, diagonal)

23. Comparison of correlation coefficients Z-statistic

24. Comparation of correlation coefficients By simply comparing correlation coefficients, no one formula seems superior to another one, in predicting visual assessment Then to establish if statistical differences exist, Z(n) is calculated : where r 1 and r 2 are the correlation coefficients to compare, n 1 and n 2 are the number of pairs, in this case equal. Standard Statistical Tables For n >120 Confidence levelsZ-statistic value 0,600,842 0,801,282 0,901,645 0,951,960 0,992,576 If the Z value obtained is greater than Z statistic, then the two correlation coefficients are statistically different

25. Table 4. Comparison of correlation coefficients Formular1Formular2CommentConfidence level DECIE20000,798LCD0,678Significantly different90% DECIE20000,798CMC0,792Not significantly different>60% DECIE20000,798CIELAB0,705Significantly different80% DECIE20000,798CIE940,771 Not significantly different>60% DECMC(2:1)0,792CIELAB0,705Significantly different80% DLNo significant difference>60% DCNo significant difference>60% DHCIE20000,777CIELAB0,709Significantly different70% It seems that only CIELAB is different to other formulae If Z value obtained is greater than Z statistic, then the two correlation coefficients are statistically different

26. Comparison in the blue region The correlation coefficient for CIEDE2000 seems the best, but does not pass the Z-statistic test The correlation coefficient for CIEDE2000 seems the best, but does not pass the Z-statistic test LCD formula is an exception, as in the previous comparison LCD formula is an exception, as in the previous comparison Table 4. Results of linear regressions in a specific area of CIE a*b*diagram (blue shades) CORRELATION COEFFICIENTSDEcolorimetric-DEvisual SAMPLES OF THE BLUE REGION36 pairs h ranging between 200º and 290º rComparison with DECIE2000-DEVIS CIELAB0,8014Not significantly different CMC210,8044Not significantly different CIE940,7767Not significantly different LCD0,6247Significantly different at a confidence level of 70% CIE20000,8108

27. Performance factors PF/4 and PF’/3 values (Luo, Kim and coll.) increase as disagreement between DV(vis) and DE(col) increases

28. Table 6. PERFORMANCE FACTORS Number of pairs: 106 TOTAL COLOUR DIFFERENCE, DE CIELAB CMC(2:1) CIE94LCDCIEDE2000 PF/439,8 31,2 34,451,8 30,8 PF’/343,2 34,7 38,358,3 34,4 BLUE SAMPLES (h between 200º and 290º) Number of pairs: 36 CIELAB CMC(2:1) CIE94LCDCIEDE2000 PF/434,7 30,1 34,450,4 30,8 PF’/339,3 33,7 38,155,2 34,4

29. Performance factors New CIEDE2000 formula gives best performance for DE New CIEDE2000 formula gives best performance for DE In the blue region it seems that CIEDE2000 is not the best option In the blue region it seems that CIEDE2000 is not the best option In the blue region, CMC(2:1) performs the best of the formulae tested, but with small differences versus CIEDE2000 In the blue region, CMC(2:1) performs the best of the formulae tested, but with small differences versus CIEDE2000

30. Conclusions Old 1995 study showed that CMC (2:1) correlates better than CIELab with visual assessments (in textile samples) Old 1995 study showed that CMC (2:1) correlates better than CIELab with visual assessments (in textile samples) New 2005 study has not found statistically significant differences in the performance between CIEDE2000, CMC(2:1) and CIE94, both in the whole set and the “blue” subset New 2005 study has not found statistically significant differences in the performance between CIEDE2000, CMC(2:1) and CIE94, both in the whole set and the “blue” subset The correlation coefficient is marginally higher for CIEDE2000 than for CMC(2:1) The correlation coefficient is marginally higher for CIEDE2000 than for CMC(2:1) No difference in Z-statistic, nor in PF values, when comparisons are made between the two formulae No difference in Z-statistic, nor in PF values, when comparisons are made between the two formulae Best PF values for CIEDE2000 in total set Best PF values for CIEDE2000 in total set Best PF values for CMC (2:1) in “blue”subset Best PF values for CMC (2:1) in “blue”subset

31. Remarks The conclusions of this study should be considered as a little contribution to the implementation of the new CIEDE2000 colour difference formula. New trials in other countries would be welcome. The conclusions of this study should be considered as a little contribution to the implementation of the new CIEDE2000 colour difference formula. New trials in other countries would be welcome. As it was agreed in ISO/TC38/SC1/WG7 meeting last year in Terrassa, up to the moment there is no objective reason to move from CMC(2:1) in ISO 105 J03 Standard, as this formula is well known and widely used in pass-fail work. As it was agreed in ISO/TC38/SC1/WG7 meeting last year in Terrassa, up to the moment there is no objective reason to move from CMC(2:1) in ISO 105 J03 Standard, as this formula is well known and widely used in pass-fail work.

32. Acknowledgements Prof. Gibert and myself would like to thank all the other contributors to this study, Mr. Dagà, Prof. Gilabert, and especially all the members of Colour Measurement Group: Mrs. Abad, Mr. Aguilar, Mr. Carmona, Dr. Enrich, Mrs. Guerrero and Marco, Mrs. Luria, Mrs. Manau, Mrs. Pérez, Mrs. Sitjà, Mrs. Roig, and the other observers Mr. Martínez (ETSEIT) and Nuria and Jordi, daughter and son of Mr. Gibert. We are gratefully indebted to the DEK committee for his kind invitation to this successful Congress at Erding in the occasion of his 50th anniversary.