Disordered spatial pattern of redness signal on facial skin and visual perception of health, stress, and hidden aging

Abstract

Background

Well‐being is commonly communicated across industries; however, experimental understanding how human perceive skin health and skin stresses are not sufficient.

Materials and Methods

Image analysis algorithm, a* gradient, was developed to evaluate spatial pattern and shape of red signal on skin. Human perception for skin health and stresses were compared with technical measurements in two visual perception studies.

Results

a* gradient correlated with perceived Inflamed Skin (R = 0.73, p < 0.01), Stressed Skin (R = 0.79, p < 0.01), Sensitive Skin (R = 0.75, p < 0.01), Healthy Skin (R = ‐0.83, p < 0.01), and Start Aging (R = 0.75, p < 0.01).

Conclusions

Disordered spatial pattern of redness signal drives human perception of skin health, stress, and aging. This new skin index of redness signal shows higher correlation with those human perception than basal a* mean, unevenness of a*, and other conventional skin color attributes.

1. INTRODUCTION

Well‐being is a focal point across industries, and as such, it is imperative not only to maintain biologically healthy skin conditions, but also to maintain healthy and youthful looking skin.

Numerous studies have been conducted to understand how facial shape and the symmetry impact on human perceptions for health, attractiveness, and age. ¹ , ² , ³ , ⁴ On top of those overall facial shape, it has been revealed that appearances of facial skin, particularly skin color, and texture, significantly influence these human perceptions. ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ Through those researches, overall color (L*, a*, b*) on cheek and the evenness and homogeneity are tested and confirmed their correlation with human perception; however, some inconsistent observations were reported especially on redness (a*) due to different study design to evaluate those perception. In the study manipulating overall a* value on the same facial image, increased a* enhanced health appearance. ¹¹ On the other hand, in the study comparing among subjects with different skin redness, no clear correlation was observed between a* and skin health perception. ¹⁰ These incongruent results imply that there are some color attributes which were not well explained in those statistical values driving human perception for skin health.

Skin redness is dominantly dependent on the concentration and distribution of hemoglobin. ¹² Decreased blood perfusion is associated with actual health condition, such as anemia, therefore skin redness is one of the indicators of oxygenated blood condition driving healthy perception. ¹¹ , ¹³ Conversely, localized red signals are associated with dermatological disease, such as Rosacea inducing transient or persistent erythema and visible blood vessels. ¹⁴ , ¹⁵ Apart from severe cases, daily environmental stressors such as UV, dryness, and chemical and bacterial infection, trigger inflammation, resulting in temporary and chronic red signals by altering vascular permeability and inducing angiogenesis. ¹⁶ , ¹⁷ As mathematical models explain redness patterns caused by disordered vascular permeability and angiogenesis, ¹⁸ , ¹⁹ there are certain shape and spatial pattern of redness generated by acute and chronic inflammation; however, association between human perception and shape and spatial pattern of redness (a*) has not been well studied.

We hypothesized that skin with red signal localizing in uneven spatial pattern and shape is perceived more unhealthy because of some reasons, (1) increasing a* value evenly on entire facial skin enhanced healthy appearance ¹¹ , (2) temporal inflammation is induced by secretion of inflammatory mediators result in temporal circular erythema, ¹⁸ and (3) semi‐chronic inflammation and cellular senescence followed by angiogenesis causes complex and uneven pattern of redness. ¹⁹ , ²⁰ Therefore, two rating study are performed to examine this hypothesis.

2. MATERIALS AND METHODS

2.1. Collecting facial skin images

To collect facial skin images, total 171 healthy Chinese female aged between 20 and 50 participated the main study. Facial skin images collected from 84 subjects aged between 20 and 39 (mean = 29.8 (SD = 5.7)) were used for the rating study. Prior to capturing facial images, subjects washed their faces and then underwent 30 min of acclimation. Full face 2D images were captured by Visia‐CR imaging system (Canfield Scientific, USA). Visia‐CR is a commercialized clinical imaging device. ²¹ , ²² VISIA‐CR can capture entire face area in various light modalities, and we used standard 2 (S2) modality for visual grading studies and cross‐polarized (XP) modality for image analysis. Captured images were saved in 3648 × 5472 pixel in BMP format. Data collection and skin acclimation occurred in the facility at 21 ± 1°C and 45 ± 5% relative humidity. Written and witnessed informed consent explaining objectives, risks, and benefits of the study was obtained from each subject upon the study enrolment. The study protocol was approved by the Ethical Committee of Global Product Stewardship in P&G International Operations (SA) Singapore Branch (ethical approval number SG21‐0018).

2.2. Skin image manipulation

To evaluate human perception difference driven by shape and pattern of red signal on skin, a facial skin image was manipulated to generate seven images having different red signal localization shape and pattern with the same L*, a*, b* mean and SD. First, a facial skin image collected in the study, was cropped into 1600 × 2250 pixel on cheek. Second, the cropped image was split into L*, a*, b* channels. Third, position of each pixel on the a* channel (1600 × 2250 pixel) was rearranged, randomly without localization (Figure 1A), localized into 1 point with controlled circular shape (Figure 1B, C), localized into 3 points with controlled circular shape (Figure 1D, E), localized into 9 points with controlled circular shape (Figure 1F) and localized into 3 points in uneven shape (Figure 1G). The manipulated a* channel image was then merged with original L* channel image and b* channel image to generate the manipulated color images.

FIGURE 1.

Manipulated images used for visual grading. (A) random distribution without localization, (B) and C) localized into 1 point with circular shape, (D and E) localized into 3 points with circular shape, (F) localized into 9 points with circular shape, (G) localized into 3 points in uneven shape.

2.3. Visual perception study with manipulated skin images

Visual perception study was conducted using computers connected to monitors (ColorEdge CG2420, EIZO Inc., CA) at Procter and Gamble in Singapore. Monitors were color calibrated using Datacolor Spyder 3 (EIZO Inc.) before the study to ensure constant color conditions and resolutions among all participants.

In this visual grading study, Chinese and Singapore Chinese female living in Singapore were recruited and a total 22 graders, aged 20–47 (mean = 33.3 (SD = 6.1)), were qualified as naïve graders. Each pair of seven manipulated skin images was randomly shown on the monitor and each grader was asked “Which skin is more inflamed?” selecting one answer from A, B, or “Same.” The total count of the selection was calculated as a rating for the inflamed skin of the image.

2.4. Visual perception study with actual facial skin images

All 84 facial skin images were cropped to 1600 × 2300 pixel on the cheek. For this visual grading study, a total 577 Japanese female living in Japan between the ages of 20 to 50 (mean = 36.3 (SD = 7.8)) was recruited as a naïve grader. Each grader accessed to online platform through their own smart phone device for the grading. Twenty image samples selected from 84 images randomly were displayed on screen in a sequence in random order. Graders were requested to rate one of five questions in one session; “Stress skin,” “Inflamed skin,” “Sensitive skin,” “Healthy skin,” “Start aging,” using 6‐point rating (1 = strongly disagree, 6 = strongly agree). Each grader participated five sessions to answer all five questions. Average scores among graders were calculated as the VPS scores of the image for each perception.

a* gradient algorithm development

An image analysis algorithm was developed to evaluate spatial pattern and shape of red signal on skin. Input facial skin image was first, converted into L*, a*, b* color space. Second, the color element and texture element were split using frequency filter on a* channel image. Third, filtered a* channel image was divided into blocks. Then, a* gradient value is calculated applying the formula.

$$\mathrm{a}\ast \mathrm{gradient}=\left(\left|\sum _i{I}_{i,j}-\sum _i{I}_{i+1,j}\right|+\left|\sum _j{I}_{i,j}-\sum _j{I}_{i,j+1}\right|\right)/A$$

where i and j are column and row value of block, I _i,j is a mean of a* channel value within a block at the position (i, j) and A is an area within ROI.

3. RESULTS

a* gradient is an indicator of uneven red signal localization on skin images

The a* gradient algorithm was developed based on the hypothesis that skin exhibiting red signals in an uneven pattern, characterized by diffused color and irregular shape, is perceived as more inflamed and unhealthy skin. This image analysis algorithm distinguished between skin images with equivalent L*, a*, b* mean values and SD, assigning higher scores to images where the red signal was localized in uneven patterns and shapes (Figure 1).

3.1. Visual perception study with manipulated skin images

To understand the impact of the shape and pattern of redness signals on facial skin and human perception of skin inflammation, seven manipulated skin images, with differing a* gradient values but identical L*, a*, b* mean and SD, were evaluated in a visual perception study. There was a strong positive correlation between a gradient value and perception for inflamed skin (Pearson correlation coefficient R = 0.93, p < 0.01) (Figure 2), that is, images with more diffused color and uneven shape of localized red signal, leading to a higher a* gradient value, were perceived more inflamed.

FIGURE 2.

Scatterplot with a regression line of association between visual rating score for inflamed skin and a* gradient. Pearson correlation coefficient and p value from the significance test are shown in the graph. Error bar shows standard deviation of total count of selection among 22 graders.

3.2. Visual perception study with facial skin images from different subjects

In the visual perception study using actual facial skin images from various subjects, strong correlations of a* gradient and human perceptions were identified: Inflamed Skin (R = 0.73, p < 0.01), Stressed Skin (R = 0.79, p < 0.01), Sensitive Skin (R = 0.75, p < 0.01), Healthy Skin (R = −0.83, p < 0.01), and Start Aging (R = 0.75, p < 0.01) (Figure 3, Table 1). These correlations observed with a* gradient surpassed those of other skin color attributes; L*, a*, b* mean, SD and color homogeneity which have been commonly utilized in prior studies for skin evaluation. ⁸ , ⁹ , ¹⁰ , ¹⁴ 65 pixel of block size was used for homogeneity analysis on original VISIA images (3658 × 5472 pixel).

FIGURE 3.

Scatterplot with a regression line of association between visual rating score and a* gradient. Pearson correlation coefficient and p value from the significance test are summarized in Table 1.

TABLE 1.

Pearson correlation coefficient of visual rating score with image analysis output values.

	Inflamed skin	Stressed skin	Sensitive skin	Healthy skin	Start aging
a* gradient	0.73	0.79	0.75	−0.83	0.75
	(<0.01)	(<0.01)	(<0.01)	(<0.01)	(<0.01)
L* mean	−0.40	−0.43	−0.35	0.51	−0.47
	(<0.01)	(<0.01)	(<0.01)	(<0.01)	(<0.01)
a* mean	0.52	0.54	0.51	−0.60	0.52
	(<0.01)	(<0.01)	(<0.01)	(<0.01)	(<0.01)
b* mean	0.00	0.02	−0.09	−0.08	0.16
	(0.99)	(0.88)	(0.41)	(0.46)	(0.14)
L* SD	−0.08	−0.12	−0.11	0.10	−0.21
	(0.49)	(0.29)	(0.32)	(0.38)	(0.05)
a* SD	0.64	0.62	0.67	−0.60	0.43
	(<0.01)	(<0.01)	(<0.01)	(<0.01)	(<0.01)
b* SD	0.21	0.31	0.26	−0.36	0.47
	(0.05)	(<0.01)	(0.02)	(<0.01)	(<0.01)
Homogeneity	0.57	0.61	0.61	−0.60	0.53
	(<0.01)	(<0.01)	(<0.01)	(<0.01)	(<0.01)

Note: p values from the significance test are shown in brackets.

4. DISCUSSION

In this paper, a novel image analysis method, a* gradient, has been introduced to evaluate disordered shape and pattern of localized red signals on skin. With this algorithm, skin displaying minimal or no red signal is scored as a low a* gradient value, skin presenting localized red signals with uniform color and/or consistent circular shape is scored as a medium a* gradient value, and skin displaying localized red signals with uneven color and/or irregular shape, often resembling amoeba‐like forms, is scored as a high a* gradient value.

Our findings confirm that the a* gradient correlates strongly with human perception of skin inflammation, stress, and health. In previous studies, statistical values of a* (specifically a* mean and/or a* SD) or color homogeneity within a region of interest have been utilized as indicators of these perceptions, as redness localization on the skin is typically associated with skin inflammation. We found that a* gradient exhibited the strongest correlation among all tested image analysis attributes, even though a* mean, a* SD, and color homogeneity still showed significant correlations with human perceptions. These results suggest that intensity and unevenness of the red signal on skin are pivotal attributes in human perceptions of skin inflammation, skin stress, and skin health. However, these attributes alone are not sufficient to encapsulate human perception; the shape and pattern of the red signal play a crucial role, as discerned by our eyes, in differentiating these perceptions. The shape and pattern influence not just perceptions related to skin health, but also perception of aging. Specifically, skin displaying localized red signals with uneven colors and/or irregular shapes is perceived as more aged.

Overall, skin displaying localized redness with uneven coloring and irregular shapes is perceived as more inflamed and stressed than skin with distinct circular red signal, characteristic of conditions like acne and acne scars. Interestingly, skin with multiple acne spots, which scores medium to high a* gradient, is perceived as inflamed and stressed, but not as showing signs of aging. This might be attributed to the distinct, circular redness being perceived as indicative of temporary damage based on individuals’ daily life experiences, whereas redness with uneven coloring and irregular shapes is viewed as a signal of accumulated damage. It remains a question as to how these patterns and shapes of red signals correlate with the skin's biological conditions.

5. CONCLUSION

A novel image analysis method has been developed and validated to assess localized red signals on the skin. The method provides a comprehensive assessment of localized red signals on the skin, encompassing not just signal intensity, but also its shape and pattern, aligning with human perceptions of skin inflammation, stress, and health. It can be applied to predict human perceptions and evaluate the efficacy of skin treatments.

Omotezako T, Neo E, Zhu H, Eharman M. Disordered spatial pattern of redness signal on facial skin and visual perception of health, stress, and hidden aging. Skin Res Technol. 2024;e13628. 10.1111/srt.13628

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.