The purpose of this research is to clarify the effect of the surrounding condition and the magnitude of the color change on the perceptual color change for the complex image. The collected data were used to optimize the parameters for the color difference equations.
- Aim of the Investigation
- Thesis Outline
In Chapter 6, the result of the main experiment is presented together with the proposed, new color change equations. The main experiment was designed to investigate the effect of ambient illuminance and color change magnitude on perceptual color change.
The color difference equations and the color appearance model are introduced, and the color difference equations were used for the analysis of data from the experiment.
- Trichromatic theory
- Opponent-colour theory
The visual system controls the sensitivity of cone cells to enable perception when the color of illumination changes (Mullen, 1985). Hunt conducted an experiment to investigate the effect of adaptation condition on color perception (Hunt, 1952).
Psychophysical Experiment Technique
- Threshold techniques
- Matching techniques
In the case of the ascending series, an observer must answer “no” until a stimulus appears different from the previous stimulus. The other is that the standard stimulus is not used and observers are asked to measure the magnitude of the stimuli based on their experience.
- CIE Color matching function
- Tristimulus values
- Colour appearance attribute
- Visual Field
- Uniform colour space
The field of view definition was used to develop a color appearance model. The uniform color space proposed by the CIE can be one of the models of color appearance, as it compares the stimulus to a reference white.
Colour difference equation
It is possible to combine the difference of each color axis to calculate the total color difference using the Euclidean distance. It is one of the standards to evaluate small color differences in the dye industry. It is examined based on the available visual data and evaluated color difference equations for use in industry.
It is relatively simpler than CMC, but the performance of CIE94 is significantly better than CIELAB color difference. CIE recommends some color difference formulas to estimate color difference for complex images (CIE, 2011b).
Image difference equation
- Mean square error (MSE)
- Signal to noise ratio (SNR)
- Peak Signal to Noise Ratio (PSNR)
- Universal Image Quality Index (UIQ)
- Structural Similarity index (SSIM)
- Hong and Luo Hue angle algorithm
- Spatial Hue Angle MEtric (SHAME)
The optimized parameter for the test image set is CIEDE and it has the best performance among the color difference equations, CIELAB ΔE*ab, CMC(l:c), CIE94 and CIEDE00. It uses the definition of the structural information in an image to calculate the index. The total color difference is estimated by applying a weight to the sum of the color difference between two adjacent pixels.
For each hue angle present, the average color difference of all pixels with the same hue angle in the image is calculated and stored in CD[hue]. They used adaptive weighting factors depending on the large area of the same color, the different hue angles and the large color difference between two adjacent pixels.
The experiment was designed to clarify the effect of ambient condition and size of color difference on perceptual image difference. Data collected through the psychophysical experiment were used to develop new image color difference equations for different ambient conditions. Before conducting the main experiment, a pilot test was conducted to investigate the effect of color difference magnitude on perceptual image difference.
In the main experiment, the effect of surround mode and color difference size on perceptual color difference, it was investigated how perceptual color difference changes for changes in surround luminance level and color difference size. The method of the psychophysical experiment was similar to that of the pilot test, which evaluated the color difference between two images (Phase.
- Uniformity test
- Colour gamut and Tone curve
- Characterization modelling
- Performance test
- Stability and Repeatability
- Characterization modelling
It has three types of light sources, D65, A and F, and the arrangement of the light sources is shown in Figure 3-9 (a). The same white spot was used (the white spot on the X-rite GratagMacbeth chart) for the stability test. The repeatability test was evaluated the next day and the following week of the first measurement day.
The lighting system was characterized by measuring every 20 steps of the input signal. For each step, the luminance of the white area of the GratagMacbeth card was measured every 5 minutes; this was continued for 10 minutes.
Surround Luminance Setting
The change in displayed color gamut is measured by changing surround conditions, as the glare reflected from the screen is affected by the level of surround luminance.
As a result, the image set for the pilot test had 195 images including 13 reference images and the image set for the main experiment had 2100 test images including 12 reference images. The size reference pair estimate for each stage and pilot test is shown in Figure 3-19 with the mean color difference ΔE*ab and ΔJ, ΔC and Δh. When lightness was manipulated and others fixed in CIECAM02, the mean difference for hue and hue was 0.42 and 1.72 for the pilot test image set and 0.39 and 1.09 for the main test image set.
The average lightness and hue change in CIECAM02 were 0.14 and 3.66 for the pilot test image set and 0.08 and 1.43 for the main experiment image set while chroma was controlled. The difference of lightness and chroma in CIECAM02 was 0.13 and 0.36 for the pilot test and 0.04 and 0.27 for the main experiment while hue was only controlled.
- Steps for experiment
The reference pair was shown first and the perceptual color difference is assigned as 80, and then the rest of the test stimuli were rated compared to the assigned color difference of the reference pair. Then, the reference pair was assigned the value 60 to evaluate 500 randomized stimuli for image color difference. Then, the reference pair was shown again to remember the difference before showing 15 stimuli repeatedly.
Then the reference pair for phase 1 was given again and they were requested to memorize the size of the reference pair's color difference from phase 1. They assessed the color difference of new reference pair compared to the reference pair for phase 1.
Method for data analysis
In this study, it is used to test the performance of color difference equations such as CMC(l:c), CIEDE2000, and CIE ΔE*ab against visual rating data. In this study, CV was used to compare the performance of color difference comparisons. N is the number of samples and ∆ indicates visual assessment data and ∆ is the predicted color difference by a color difference equation.
The image set for the pilot test was used to investigate the effect of color difference size on the performance of the color difference comparison. The performance of the color difference comparisons was examined for each group and all the images.
- Observer performance in pilot test
- Observer performance in main experiment
Short-term repeatability was tested using data assessed twice in the same session in dark conditions for the consistency of the repeated responses. Long-term repeatability used the test result for two stimuli shown in two sessions for the whole experiment's consistency. The single component controlled images were used for the repeatability test including the outlier test.
Outliers in Phase 1 were excluded for further analysis such as short-term and long-term reproducibility tests. Rating data from 22 participants for the dark, medium, and light conditions were used for further analysis.
Experimental Result of Pilot test
- Performance test for colour difference equations
- Parametric Factors Optimization for CMC(l:c) and CIEDE2000
The performance test for color differences was performed with one image set for the pilot test. Filled circle indicates color difference for chroma change, empty square is the color difference for hue change, and empty triangle indicates the color difference for lightness change. There are studies that investigated the performance of color differences with larger parametric factors.
The magnitudes of the overall color differences in the images must have affected the operation of the color difference equations, as in the cases of color differences for uniform color spots. This means that the optimal parametric factors are affected by the magnitudes of the color difference data.
Experimental result for Main experiment
- Effect of parametric factors, k L and k C
- The effect of surround condition on colour difference
- The effect of CIEDE2000[1:1:1] magnitude on colour difference
- Optimization result of parametric factors, k L and k C
The color difference equation predicts changes in brightness that are greater than the visual data, as shown in Figure 6-1. On the other hand, the observed color difference for L-CD in dark conditions is greater than for other environmental conditions. As the difference between the surround luminance decreases, the perceived color difference between that of the surround luminance decreases.
Therefore, the effect of ambient lighting must be taken into account for predicting color change. The comparison showed that the perceptual color difference varies depending on the magnitude of the color change and the brightness of the surroundings.
Modelling for surround-adaptive colour difference metric
- Modelling for lightness parametric factor, k L _ SOO
- Modelling for chroma parametric factor, k C _ SOO
- Steps for using New Image colour difference equation, Soo’s model DE00 SOO
- Statistical Significance test
The effect of spatial conditions and the magnitude of color differences on parametric brightness factors is significant and there is a relationship between the two variables. The parametric brightness factor function can be over-predicted for larger magnitudes of color difference under dark conditions, as it is a 2nd polynomial function. Furthermore, the proposed model is marginally better than CIEDE2000 [optimized parametric factors] for each group of color differences.
In this chapter, a new color difference metric was developed with functions of parametric factors for lightness and chroma. Moreover, the performance of DE00_SOO is almost the same as the optimized result for each color difference group.
Conclusions and Discussion
New image color difference equations for complex images were developed based on data from a psychophysical experiment. This means that the optimal parametric factors are affected by the magnitudes of the color difference data. The comparison showed that the changes in perceived color difference depend on the size of the color difference and spatial luminance.
Furthermore, this result means that the surround mode must be taken into account when evaluating the image color difference. It factors in the surround luminance ratio and the size of the color difference to determine the exact color difference.
The new equation can be used to evaluate the overall color difference in images such as the duplicate displays for tile display and a perceptibility test among displays produced with the same process in the luminance of the office environment and not in the dark. Furthermore, the prediction of chroma change in CAM02-UCS is limited by the use of chromaticity, M, instead of chroma, C. Although the scale between lightness J, chroma C, and hue angle h is different from others, the performance of Euclidean distance in CIECAM02 JCh outperforms CAM02-UCS in terms of chroma change.
Meanwhile, CIECAM02 has the variables to predict the appearance under various ambient conditions, the calculation of color difference in CIECAM02 or the use of CAM02-UCS for the image color difference evaluation is not proper since CIECAM02 is not developed to calculate color difference, but to calculate color appearance to predict and it was also developed based on color spots not complex images. The performance of new image color difference comparisons is the best in three color difference comparisons.
Limitations and Further work
Paper presented at the Second International Workshop on Video Processing and Quality Metrics, Scottsdale, USA. Paper presented at the IS&T/SPIE 1994 International Symposium on Electronic Imaging: Science and Technology.
Visual and Optimized data