Figure 1:(Color online) Color calibration of the color sensor module. (a) Normalized sensor output at the i-channel photodiode Xi = Vi=Vi;- max (i=r, g, and b) as a function of the RGB color coordinates of i-color light from a digital projector. The green and blue symbols at the r-channel photodiode (left) represent the data measured with this photodiode under green and blue light illumination, respectively. The dashed curves are the third-order polynomial fit of the color coordinates to the measured data Xi. The insets at the middle section show the spectral profiles of the red (R=255), green (G=255), and blue (B=255) light used, which yield Vi;max. (b) Post-calibration output at the i-channel photodide. The cross-talk in the calibrated data is closer to zero than that in (a).

Figure 2:(Color online) Breathing-induced color coordinate variations. (a) RGB color coordinates (at the palm of a healthy 26-year-old Asian male) determined from the DC output channel (0–1.6 Hz) of the color sensor module. Small ripples visible on the breathing signal envelopes are attributable to heartbeats. (b) Corresponding PCA outputs of the RGB traces shown in (a) by removing the skin tone (R = 153, G = 117, and B = 107). (c) CIELAB color coordinate traces are relevant to (a) comprising lightness L*, the red/green opposing colors a*, and the yellow/blue opposing colors b*. (d) Corresponding PCA outputs of the L*a*b* traces shown in (c).

Figure 3: (Color online) Heartbeat-induced color coordinate variations. (a) RGB color coordinates at the palm of a healthy 26-year-old Asian male retrieved from the AC channel output (0.1–28 Hz) of the color sensor module. (b) First three PCA components of RGB traces shown in (a). (c) CIELAB color coordinate traces relevant to (a), comprising lightness L*, the red/green opposing colors a*, and the yellow/blue opposing colors b*. (d) First three PCA components of (c).

Figure 4:Fast Fourier transform plot of heartbeat-induced color coordinate variations. (a) Measured RGB signals (R: top, G: middle, B: bottom) at the AC output of the color sensor module measured at the palm of a healthy 26-year-old Asian male. ECG traces (black) are plotted as the timing reference. (b) Fourier transform of the measured AC signals shown in (a). (c) RGB signal traces shown in (a) are expressed in the L*a*b* color coordinates (L*: top, a*: middle, b*: bottom). (d) Fourier transform of the L*a*b* color coordinate signals shown in (c).

Figure 5: Gaussian mixture model fitting of breathing-induced color variation data. A 120-min PC1 signal trace from the DC output (0.1–28 Hz) of the color sensor module is fitted to the GMM of three states (S1 coded by red, S2 by green, and S3 by blue). The color signals are measured at the palm of a subject at rest (left) and after physical exercise (right).

Figure 6: Gaussian mixture model fitting of heartbeat-induced color variation data. A 120-min PC1 signal trace at heartbeat rhythm from the AC output of the color sensor module is fitted to the GMM of three states (S1 coded by red, S2 by green, and S3 by blue). The color signals are measured at the palm of a subject at rest (left) and after physical exercise (right).

Figure 7: Relaxation trajectory of heartbeat-induced color variation after physical exercise. A 120-min-long PC1 sequence is divided into a series of 5-min-long segments. aj and ?tj extracted from each interval are fitted to a Gaussian distribution profile and the resultant contours are coded from red to green. Relaxation trajectory is displayed by connecting the centers (blue star symbols) of each contour profile with lines. The Gaussian distribution profile at rest is characterized before physical exercise and is presented in black.

Figure 8: Hidden Markov model fitting of heartbeat-induced color variation data. (a) A 360-min PC1 signal trace taken from the AC output of the color sensor module is fitted to the hidden Markov model with three hidden states (S1 coded as green, S2 as red, and S3 as blue). The black line is a projection of the data points. (b) Histograms of the three hidden states along the black projection line. (c) The inline distances of 4,000 data sequences are plotted against the times of occurrence to reveal the transition kinetics. (d) The hidden state transition diagram collected from a 360-min-long sequence. This figure has been revised from ref. [26] to make the discussion of this paper more complete.

Figure 9: Physiological states for different health conditions. Gaussian distribution profile of 30-min heartbeat-induced color variations (expressed in the L*a*b* coordinate system) in 26-year-old healthy Asian males: one with a habit of daily physical exercise (black), one with a habit of irregular exercise (red), and one with a habit of no exercise (blue).

Figure 10: Physiological states for different ages. Gaussian distribution profile of 30-min heartbeat-induced color variations in a 26-year-old (black) and a 65-year-old (pink) Asian male.