Prediction of Color Appearance Change of Digital Images under Different Lighting Conditions Based on Visible Spectral Data Ken-ichiro Suehara, Makoto Hashimoto, Takaharu Kameoka and Atsushi Hashimoto Division of Sustainable Resource Sciences, Graduate School of Bioresources, Mie University, Tsu, Japan Tel: +81-9-231-963; E-mail: hasimoto@bio.mie-u.ac.jp Abstract For the final purpose of the continuous remote monitoring of the surface color changes of crops and agricultural products with the atmosphere condition data during cultivation, color appearance changes of digital images taken under different lighting conditions were investigated to construct a color calibration method based on the spectral feature information of the lighting. The images of the standard color chart with the plain Red, Green and Blue sections were taken using digital camera and RGB values of the color appearance change were measured in the standard color acquisition system. The visible spectral radiance of the lighting ranging from 38 to 73 nm in 1 nm intervals was also measured. Multiple linear regression analysis (MLR) was carried out between the spectral feature information and the color appearance change of the chart image. As the results, good correlation was observed. Prediction of color appearance change of the standard color chart image was possible based on the spectral feature information of the lighting. Construction of color calibration method of digital image without using the standard color chart may be possible. Keywords: standard color chart, color calibration, visible spectrum, lightning, field monitoring Introduction The color of agricultural products could be reflected by the pigment components and the geometrical structure and is an important quality factor. In particular, the color is used as a useful quality parameter at cultivation, harvesting, sorting and packing. Various machine vision studies on the color of agricultural products have been reported. Generally, color parameters in the L*a*b* color system or in the RGB color coordinate system are used for the color evaluation of agricultural products. We had adopted the HSL (H: Hue, S: Saturation, L: Lightness) color system based on the CCD camera since the color measurement using CCD camera had the advantage that it easily acquired the surface information and the color distribution (Kameoka et al., 1994; Motonaga et al., 1997). And we developed the continuous and quantitative remote monitoring method of the surface color changes of agricultural products with the atmosphere condition data during the cultivation. Surface color change of tomato fruit on the tree could be remotely monitored through the maturing process using the Field Server (Fukatsu and Hirafuji, 2) equipped with a digital camera (Hashimoto et al., 26). IAALD AFITA WCCA28 181
Change in the lighting condition affects the color appearance of recorded digital image. We developed the color calibration method of the image using standard color chart (Hashimoto et al. 21). In this method, standard color chart was taken within the same image area of the subject and true surface color of fruit was measured by using the chart color. In the practical situation, however, it is unacceptable to put always the standard color chart at the side of the subjects (agricultural product, crops of wide area in the field, scenery of hinterland and pole etc.) and the chart will discolor under the field condition. Therefore, development of color calibration method without using the standard color chart should be necessary. The present purpose is to develop a color calibration method with the spectral data of the lighting and to construct the method without using the standard color chart in the future. As the first step of the study, prediction of color appearance change of standard color chart image under different lighting condition based on spectral data of lighting was investigated using the standard color acquisition system. Materials and methods The images of the standard color chart were taken using a single-lens reflex camera with 6.17 million-pixel CCD (FinePix S3Pro, Fuji Photo Film), and were recorded as RAW format files. The image acquisition was carried out in the standard color image acquisition system (Motonaga et al., 24) equipped with the digital camera, two fluorescent light at K color temperature (TRUE-LITE, DURO-TEST Co., Ltd.) that is based on the CIE (Commission Internationalede l'eclairage) regulations, and diffuse reflectors (Fig. 1). Illuminance of the surface of the sample stage was adjusted to 14 lx by changing the angle of the diffusion reflectors. Dark room Digital camera Diffuse reflector Fluorescent light Sample stage Fig. 1 Scheme of standard color image acquisition system To prepare the different lighting conditions (different spectral patterns of lighting), two kinds of fluorescent lights were used and nine color films were wound around the fluorescent lights. The images of the standard color chart put on the center of the sample stage were taken under the different lighting and the average of the R, G and B values in the RGB color space at each pixel in the square area of the R, G and B chart were obtained. IAALD AFITA WCCA28 182
Spectral radiance [mw/nm/m 2 ] [mw/nm/lm] The color appearance change was calculated based on standard lighting and the change values, R-ΔR, R-ΔG, R-ΔB, G-ΔR, G-ΔG, G-ΔB, B-ΔR, B-ΔG, B-ΔB, were measured (Fig. 2). In the figure, R R, S and R R, T mean the R value in the RGB color space of the R Chart image area that were taken under standard and test lighting. The visible spectral radiance of the lighting ranging from 38 to 73 nm in 1 nm intervals was measured with spectral measurement device (Eye-One Prophoto, GretagMacbeth Co.). The spectral values were divided by values of illuminance (14 lx) for normalization and the spectral differences were calculated by subtracting the spectral values of standard lighting at each wavelength (Fig.3). Seventy-two data sets were prepared and were divided into the one for calibration data set (n=6) and one of validation sample set (n=12). Fig. 2 Definition of color appearance change values based on standard lighting condition 6 4 3 2 1 Standard lighting Test lighting (A) 4 4 6 6 7 wavelength λ [nm] Normalized spectral differences, Iλ.2.1.1.. -. -.1 -.1 Standard lighting Test lighting (B) 4 4 6 6 7 Wavelength λ [nm] Fig. 3 Spectral radiance of the light sources ranging from 38 to 73 nm in 1 nm intervals (A) and the difference spectra for test lighting data minus standard lighting data at each wavelength after normalization (B) IAALD AFITA WCCA28 183
Results and Discussion Multiple linear regression analysis Multiple linear regression analysis (MLR) was carried out between 36 points spectral data as an explanatory variables and color appearance change values, R-ΔR, R-ΔG, R-ΔB, G-ΔR, G-ΔG, G-ΔB, B-ΔR, B-ΔG, B-ΔB as an explained variables. As the results, following nine MLR equations for prediction of color appearance change value were obtained. R (1) R ar R, I R (2) G ar G, I B B ab B, I (9) Here, a λ and I λ are MLR coefficients and values of normalized spectral differences at each wavelength,λ nm. As the results of MLR analysis, the correlation coefficients were observed over.998 in all cases. This result suggested that change of the spectral values of the lighting is reflected with the color appearance changes of the recorded digital image. Therefore, visible spectral feature information is useful to construct color calibration method for recorded digital image without using the standard color chart. Validation of obtained nine equations To validate the obtained nine equations, the values of color appearance change in the validation data set which was not used to get the equations were predicted (Fig. 4). Calibration data set (n=6) Measured Values of spectral R-Δ R,Δ G,Δ B Differences G-Δ R,Δ G,Δ B I 38, I 39,, I 73 B-Δ R,Δ G,Δ B MLR Analysis Validation data set (n=12) Measured Values of spectral R-Δ R,Δ G,Δ B Differences G-Δ R,Δ G,Δ B I 38, I 39,, I 73 B-Δ R,Δ G,Δ B Correlation? Nine equations R-Δ R = Σ a R-Δ R λ I λ R-Δ G = a R-Δ G λ I λ B-Δ B = Σ Σ a B-Δ B λ I λ Fig. 4 Schematic of MLR analysis and validation of equations Predicted R-Δ R,Δ G,Δ B G-Δ R,Δ G,Δ B B-Δ R,Δ G,Δ B IAALD AFITA WCCA28 184
The results are shown in Fig.. The actual values, R-ΔR, R-ΔG,, B-ΔB, are plotted on the horizontal axis and the values predicted by nine MLR equations, R-ΔR, R-ΔG,, B-ΔB are plotted on the vertical axis. Good correlations were obtained between the actual values and predicted values because the multiple correlation coefficients (r 2 ) were observed over.99 in all cases. Although the predicted results forδb were slightly lower than other cases, nine equations has satisfactory precision. Prediction of the color appearance change of digital image of the standard color chart with the plain Red, Green and Blue sections may be possible. And it means that an imaginary color chart image can be built using visible spectral data in computer. 6 4 2 R-Δ R r 2 =.9998-2 -2 2 4 6 1 - -1 Actual values[-] G-Δ R r 2 =.9986-1 -1-1 - 1 1 - -1 B-Δ R r 2 =.998-1 -1-1 - 1 1 1 - R-Δ G r 2 =.9986-1 -1-1 1 1-1 -2-3 Actual values[-] G-Δ G r 2 =.9996-4 -4-3 -2-1 1 1-1 -2 B-Δ G r 2 =.999-2 -1 1 1 - -1 R-Δ B r 2 =.996-1 -1-1 - 1 1-1 -2 G-Δ B r 2 =.9979-3 -3-2 -1 1-2 -4 2 B-Δ B r 2 =.9982-4 -2 2 Fig. Correlation between actual values of color appearance change and the values predicted by MLR equations IAALD AFITA WCCA28 18
Conclusion Prediction of color appearance change of the standard color chart that was acquired under condition of the different lighting was possible based on the visible spectral data. Construction of color compensation system may be possible without using the standard color chart by measuring the visible spectrum data of the lighting. However, there are tested on the ideal lighting conditions and on the standard color acquisition system in this study. Although an experiment under the actual lighting condition may be necessary, the purpose of the first step of this study could be accomplished. Prediction of the values of the color appearance changes in the image that acquired under condition of the sunlight is now being studied. References Fukatsu, T., and M. Hirafuji. 2. Field Monitoring Using Sensor-Nodes with a Web Server. J. Robotics Mechatronics. 17: 164-172. Hashimoto, A., H. Kondou, Y. Motonaga, H. Kitamura, K. Nakanishi, and T. Kmeoka. 21. Evaqluation of Tree Vigor by Digital Camera Based on Fruit Color and Leaf Shape. Proc. of the World Congress of computers in Agriculture and Natural Resources. pp. 7-77. Hashimoto, A., K. Yasui, M. Takahashi, S. Yonekura, T. Hirozumi, T. Mishima, R. Ito, K. Suehara, and T. Kameoka. 26. Remote Monitoring of Color of Agricultural Products in the Field Using a Digiral Camera and the Field Server. Proc. of the 4 th World congress on computer agriculture. pp. 66-71. Kameoka T., A. Hashimoto, and Y. Motonaga. 1994. Surface color measurement of agricultural products during post-post ripening. Color Forum Japan 94 Proceedings. Pp.11-14 Motonaga Y., H. Kondou, A. Hashimoto. 1997. Constructing color image processing system for managing the surface color of agricultural products. J. Japanese Soc. Agric. Machin. 9(3):13-21. Motonaga, Y., H. Kondou, A. Hashimoto, T. Kameoka. (24) A method of making digital fruit color charts for cultivation management and quality control. J. Food Agri. Environ. 2: 16-166. IAALD AFITA WCCA28 186