Eur. J. Lipid Sci. Technol.

109 (2007) 157–164 DOI

10.1002/ejlt.200600211 157

Kıvanc¸ Kılıc¸
a
b
A novel method for color determination of
Baran Onal-Ulusoy
˙

Ismail Hakkı Boyacı

b

edible oils in L*a*b* format

aBiasis Ltd. Sti., Kosgeb- A simple method that uses visible spectrophotometer data and an artificial neural net-
Hacettepe Technology work (ANN) was developed to determine edible oil color based on the L*a*b* format.
Research Center, The 100 oil samples consisted of nine pure oils, a sesame oil blend and three heated
Ankara, Turkey
oils. Binary, ternary and quaternary mixtures of these 13 oils in different ratios were
bDepartment of Food
Engineering, prepared, and absorbance values of the samples were measured in the visible region
Hacettepe University, (380–700 nm). The absorbance values at wavelengths of 416, 456, 483, 537, 611
Ankara, Turkey and 672 nm were used to train, validate and test the network. Strong correlations
between the instrumental L*a*b*DE and the estimated L*a*b*DE were found for the
2
test samples, with correlation coefficients (R ) of 0.989, 0.984, 0.996 and 0.992 for
L*, a*, b*, and DE, respectively. The effects of number and combination of the
wavelengths used for training of the ANN on the estimation capability of the network
2
for the test samples were also investigated. Although a good agreement, average R
of 0.991– 0 993 for L*a*b*, was obtained for combinations composed of three to six
2
wavelengths with 483 and 537 nm in common, the best R value was obtained when
all six wavelengths were used to train the ANN. The developed method is objective,
cost effective and simple, and allows the color measurement with a basic visible
spectrophotometer and dis-posable cuvettes.

Keywords: Oil, color, L*a*b*, spectrophotometer, artificial neural network.

1 Introduction imum absorbance of chlorophyll pigments, is missing in

most processed oils [1]. The Lovibond color method is the
Color is an important quality parameter of edible oil, both most widely used color measurement technique in the
in the refining process and in the marketplace. The edible oil industry, but it requires a Lovibond tintometer. A
edible oil industry often analyzes the oil color, either Lovibond tintometer consists of red, yellow, and blue
qualitatively or quantitatively, during the process to permanently colored glass standards. The addition of the
maintain a con-sistent quality. Oil appearance might be blue color field provides a greater degree of brightness and
an indicator of a problem having occurred during greenness than given by the Wesson color. The
blending, storage, crushing, and extraction or the spectrophotometric method, an instrumental technique,
refining process [1]. Each oil has its own characteristics eliminates the visual judgment of a user required in both the
color, primarily due to the presence of carotenoids Lovibond and Wesson methods. This method depends on
and/or chlorophyll pigments or gossypol. Therefore, the the calculation of the absorbance of four definite

Research Paper
oil color is often specified in the trade rules by various wavelengths. The results can be transformed to the
associations in different countries [2]. Lovibond or Wesson color values with a developed
equation. The method is not an alternative for both the
The American Oil Chemists’ Society (AOCS) [3] has pro-
Lovibond and Wesson color methods because of occa-
posed four official methods for the color determination of
fats and oils. These methods are Lovibond color (AOCS sional disagreement, economics, and the firm entrench-
method Cc 13e-92), Wesson color (AOCS method Cc ment of the visual procedure. However, it is still an official
13b-45), spectrophotometer color (AOCS Cc 13c-50) method as well as the only available objective method [1].
and chlorophyll color (AOCS method Cc 13d-55). The
Currently, CIE L*a*b*, XYZ, Hunter Lab, and RGB (Red,
latter method is not applicable to hydrogenated and
Green, Blue) are the alternative color models that might be
deodorized oils because 670 nm, representing the max-
used in objective oil color evaluation. L*a*b* is an
international standard for color measurements, adopted by
˙ the Commission Internationale d’Eclairage (CIE) in 1976 [4].
Correspondence: I smail Hakkı Boyacı, Hacettepe University, The L*a*b* values have been widely used in food studies.
Department of Food Engineering, Beytepe TR 06800, Ankara,
Turkey. Phone: 190 312 2977100, Fax: 190 312 2992123, e- L* is the lightness component ranging from 0 to 100. a*
mail: [email protected] (from green to red) and b* (from blue to yellow) are

 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ejlst.com

158 K. Kılıc¸ et al. Eur. J. Lipid Sci. Technol. 109 (2007) 157–164

two chromatic components ranging from 2120 to 120. common laboratory equipment and uses low-volume
The CIE L*a*b* color is device independent, providing disposable cuvettes in the visible region. An ANN was
consistent color regardless of the input or output device, used for the estimation of L*a*b* values from the absorb-
such as digital cameras, scanners, monitors and printers ance values obtained with the visible spectrophotometer,
[5]. In contrast to other color models such as RGB, XYZ and the performance of the trained network was further
and Hunter Lab, the color perception is uniform in the investigated by testing the size and nature of the input
L*a*b* space, i.e. the Euclidean distance between two matrix of the ANN.
different colors corresponds approximately to the color
perceived by the human eye [6]. Commercial and objec-
tive color measurement instruments measuring the color 2 Materials and methods
with alternative color methods are readily available and
have been used in several studies [7–10], but measure- 2.1 Oil samples
ments with such devices require large amounts of oil
samples (,10–13 mL), which is too much for food re- Refined soybean, corn, cottonseed, hazelnut, sunflower,
searches, and these devices are relatively expensive canola and olive oils, and a sesame oil blend (40%
and use expensive disposable sample cuvettes. sesame and 60% soybean oils), and riviera olive oil, a
blend of virgin and refined olive oils (free fatty acid
Organic molecules can absorb UV/visible light. As a content ,1.5%), and virgin olive oil were purchased from
result of the absorption of light, electronic transitions local markets. The sunflower, refined olive and soybean
occur, leading to a higher energy or excited state of the oils were heated continuously at 180 7C for 8, 8 and 30
mole-cule. Because of the decolorization of the electrons h, respectively, to get oil colors between the color of cot-
along the chromophore, the excited state of the molecule tonseed oil and a sesame oil blend. The 100 oil samples
is of low energy so that absorption of visible light is consisted of nine pure oils, a sesame oil blend, three
enough to give rise to the transition. The human eye heated oils, and binary, ternary, and quaternary mixtures
sees the waves that are left over from the absorption of of these oils (except canola oil) in different ratios, to get
the compounds, and the oil appears colored [10]. more oil samples containing different types and/or levels
of color pigments, were used.
Computerized image analysis systems has been devel-
oped and used successfully to measure the color of the
various solid food samples [8, 9, 11–13]. Fengxia et al. 2.2 Instrumental determination of color
[2] have attempted to develop an objective system to
meas-ure the color of various oils, depending on image The CIE L*a*b* values of the oil samples were measured
analy-ses. The image of the oil samples was captured using a Minolta Spectrophotometer CM-3600d (Minolta
with a scanner and a camera, and it was processed with Camera Co., Japan). The instrument was calibrated with
2 the zero calibration box and the white calibration plate,
soft-ware. A good agreement, R of 0.999, between the
visual Lovibond red color and the image analysis red respectively, and the white calibration plate was used as
color was reported. Since the regulations on oils a target for the calculation of total color difference (DE).
produced in China have limitations only for red readings, The oil samples were put into 13-mL solvent-resistant
the authors report-ed that it is not necessary to correlate cu-vettes (10 mm path length), and the cuvette was
their system to yellow and blue color readings. washed with HPLC grade hexane before each
measurement. The final color score of each oil sample
Recently, knowledge-based approaches such as statis-tical was obtained by averaging the scores of the duplicate
learning, fuzzy logic and artificial neural networks (ANN) measurements. DE values were calculated from the
have been applied successfully to the inspection of food L*a*b* values according to the following equation:
quality, modeling and control of various biological rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffiffiffiffiffiffiffiffiffiffi

processes. Neural networks mimic the human intelligence 2 2 2

LL a b
for objective learning [14–16]. The combination of image DE ¼ target þa target þb target
analysis and ANN brings about a powerful tool for machine
vision inspection. ANN have been utilized for classification, 2.3 Absorbance measurement with UV/visible
prediction and segmentation in quality evaluation of food spectrophotometer
products in the recent years [16–18].
Absorbance values of each oil sample were measured in
The objective of this study was to develop a method to the visible region (380–700 nm) with an Agilent 8453 UV-
measure the color of a wide range oil samples in CIE Visible Spectrophotometer (Agilent Technologies, Palo Alto,
L*a*b* format with a visible spectrophotometer, which is a CA, USA) and a disposable polystyrene sample cu-

© 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ejlst.co

m
Eur. J. Lipid Sci. Technol. 109 (2007) 157–164A new method for color determination of edible oils in L*a*b* format 159

vette (,3 mL) with a path length of 10 mm. Zero absorb- were used as input variables, and the designed network
ance was set with distilled water as a blank. was trained using training and validation data. The per-
Wavelengths were scanned at room temperature for formance of the trained network was investigated by
each sample in duplicate to check variations. comparison of the L*a*b* values determined with the
Minolta instrument and those estimated with the ANN for the
test samples. However, measuring six absorbance values at
2.4 Data selection six different wavelengths might be difficult with a basic
spectrophotometer. Therefore, training of the net-work using
Absorbance values of selected wavelengths (416, 456, lower numbers of input variables (absorbance values at
483, 537, 611 and 672 nm) and L*a*b* values of each selected wavelengths) were investigated in the second part
sample were collected in Matlab 7.0 structure (The of the study. All the possible combinations of six
Math-works, Natick, MA, USA). Both input (absorbance wavelengths were created and used individually as an input
values) and output (L*a*b* values) data were normalized matrix for training of the network.
using the Matlab function prestd. The means of
normalized data were zero and the standard deviations
of those data were 1. Input data were divided randomly 3 Results and discussion
into three groups: training (50 samples), test (25
samples) and vali-dation (25 samples). Absorbance spectra of nine pure oils, three heated oils and
a sesame oil blend, between 380 and 700 nm, are given in
Fig. 1. Absorption peaks at wavelengths of 416, 456, 483,
2.5 Training and testing the neural network 537, 611 and 672 nm were detected only for the virgin and
riviera olive oils. The absorbance values of the virgin olive
Feedforward neural network was used in this study. The oil at these six wavelengths were higher than the values for
network consists of an input matrix, hidden and output the riviera olive oil. This is not surprising because only the
layers. The hidden and output layers consisted of six and virgin olive oil part of riviera olive oil is not bleached, so it
three neurons, respectively. Logsig and Purelin were preserves its color pigments more than the refined oils. No
chosen as transfer functions for the hidden and output absorbance peaks were detected in the visible region for the
layers, respectively. The Levenberg-Marquardt back- pure oils, including refined olive oil. This is related mainly to
propagation (trainlm) training algorithm with early stop-ping the reduction of the color pigment concentration of edible
method was used for training of the network. In the first part oils during the bleaching process with activated earth.
of the study, six absorbance values, measured at six Absorbance peaks between 380 and 440 nm were detected
different wavelengths selected in a previous section, only for the three heated

Fig. 1. Absorbance spectra of nine pure oils,

three heated oils and a sesame oil blend at the
visible wavelengths. CSO, cottonseed oil;
SeOB, sesame oil blend; VOO, virgin olive oil;
HROO, heated refined olive oil; RivOO, riviera
olive oil; CO, corn oil; ROO, refined olive oil;
HZO, hazelnut oil; SBO, soybean oil; SFO,
sun-flower oil; HSFO, heated sunflower oil;
HSBO, heated soybean oil; CNO, canola oil.

 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ejlst.com

160 K. Kılıc¸ et al. Eur. J. Lipid Sci. Technol. 109 (2007) 157–164

oils and the sesame oil blend. These peaks may be Absorbance values at six wavelengths for 50 randomly
relat-ed to colored oxidation products occurring during selected oil samples were used for training the network.
heating of these three oils and during roasting of the The performance of the trained network was investigated
sesame seeds. The absorption peaks detected at 416, by comparison of the measured L*a*b*DE values with
456 and 483 nm mainly belong to isomers of carotene or the estimated L*a*b*DE color values. The relationship
chloro-phyll, because isomers of carotene and some be-tween the L*a*b*DE values measured using the
isomers of chlorophyll have been reported to have Minolta instrument (instrumental L*a*b*DE) and those
absorption peaks at 400–480 and 430–466 nm, estimated by ANN for the training data set is shown in
2
respectively [10, 19]. Since flavonoids and some isomers Fig. 2. The correlation coefficients (R ) between the two
of chlorophyll pigments have been reported to have systems were 0.995, 0.994, 0.997 and 0.996 for L*, a*,
absorption peaks at 500–550 and 670 nm, respectively, b* and DE, respectively. The plots of the instrumental
the absorption peaks at 537 and 672 nm strongly L*a*b*DE vs. the L*a*b*DE estimated with the trained
indicate the presence of color pig-ments and phenolic ANN were linear, having slopes close to unity.
compounds [1, 3, 20, 21]. Although there was no
information found on what kind of a com-pound might The trained network was validated with the absorbance
have an absorption maximum at 611 nm, Paul and Mittal values of 25 oil samples, the second data group. The
[22] have reported that major degradation products results obtained from the ANN were compared with
might have absorption peaks between 550 and 695 nm. those measured with the instrument. Strong correlations

Fig. 2. Experimentally determined L*a*b*DE values vs. estimated L*a*b*DE values for training data.

 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

www.ejlst.com
Eur. J. Lipid Sci. Technol. 109 (2007) 157–164A new method for color determination of edible oils in L*a*b* format 161

between the instrumental L*a*b*DE and the L*a*b*DE and DE, respectively. These results indicate that the
estimated with the ANN were found, with correlation ANN successfully estimated the L*a*b* values using
coefficients of 0.986, 0.980, 0.995 and 0.991 for L*, a*, b* absorbance values measured at six wavelengths.
and DE, respectively (Fig. 3). The plots of the instrumental
L*a*b*DE vs. the L*a*b*DE estimated with the ANN were Absorbance measurements at six wavelengths for one
linear and had slopes between 0.966 and 0.988. sample may be time consuming, unless a spectro-
photometer with a diode array detector is used. Chem-
The performance of the trained and validated network was ical compounds may give more than one absorption
further tested with the last group of data collected from 25 peak at different wavelengths. Measurement of absorb-
oil samples having different color intensities. The results of ance at one wavelength may give information about the
the ANN were compared with those obtained from the absorbance values at the other wavelengths. In the
2
instrument (Tab. 1). R values between the instrumental second part of the study, the possibility to train the same
L*a*b*DE and the L*a*b*DE estimated with the ANN were network using a lower number of absorbance values was
determined as 0.989, 0.984, 0.996 and 0.992 for L*, a*, b* investigated. The effect of the number of the

Fig. 3. Experimentally determined L*a*b*DE values vs. estimated L*a*b*DE values for validation data.

 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ejlst.com

162 K. Kılıc¸ et al. Eur. J. Lipid Sci. Technol. 109 (2007) 157–164

§
Tab. 1. Comparison of instrumental and ANN-estimated color measurements for various oil samples .

Pure oil/oil mixtures (wt/wt) Instrumental ANN

L* a* b* ˜E L* a* b* ˜E

CO 40.2 21.8 10.7 59.8 40.1 21.7 10.5 59.8

SFO 41.2 21.2 4.2 57.9 40.9 21.2 4.5 58.2
CSO1CO (1 : 1) 40.0 21.8 10.7 60.0 40.1 21.8 10.6 59.9
CSO1SFO (1 : 1) 40.5 21.8 8.6 59.1 40.5 21.7 8.2 59.1
VOO1ROO (1 : 1) 36.5 0.9 21.2 66.0 36.8 1.1 21.4 65.8
VOO1HSFO (1 : 1) 36.0 2.3 20.2 66.3 36.0 2.1 20.8 66.4
FROO1RivOO (1 : 1) 38.5 20.2 17.7 63.1 39.1 22.0 18.2 62.7
FROO1SBO (1 : 1) 40.0 22.8 15.5 61.1 39.6 22.6 16.3 61.7
RivOO1HZO (1 : 1) 39.9 21.5 10.7 60.1 39.9 21.6 11.2 60.2
RivOO1HSBO (1 : 1) 36.8 1.9 20.7 65.6 37.8 0.7 20.4 64.5
CO1SFO (1 : 1) 40.7 21.4 7.1 58.7 40.5 21.4 7.2 58.9
HZO1SBO (1 : 1) 41.1 21.2 3.8 58.0 41.0 20.9 4.0 58.1
SBO1SFO (1 : 1) 40.9 21.4 4.7 58.3 40.9 21.1 4.5 58.2
SeOB1SBO (3 : 2) 30.3 6.6 9.8 69.7 29.9 7.1 9.3 70.1
VOO1CO1HROO (1 : 1 : 1) 37.4 0.5 21.5 65.3 37.3 0.5 21.0 65.2
VOO1SBO1HSFO (1 : 1 : 1) 37.2 0.8 20.9 65.1 36.9 1.1 21.1 65.7
VOO1SFO1HSBO (1 : 1 : 1) 36.6 2.4 20.6 65.7 36.9 1.4 20.6 65.5
RivOO1CO1HROO (1 : 1 : 1) 39.6 22.1 16.3 61.7 39.4 22.1 16.1 61.8
RivOO1SBO1HSFO (1 : 1 : 1) 38.8 21.7 17.4 62.7 39.0 21.8 17.0 62.4
RivOO1SFO1HSFO (1 : 1 : 1) 39.2 21.7 17.3 62.3 39.0 21.8 17.0 62.4
CSO1SeOB1SBO (5 : 6 : 4) 33.7 4.3 14.3 66.8 33.7 3.8 14.3 66.9
SeOB1SBO1HZO1HROO (6 : 4 : 5 : 5) 33.5 4.5 14.6 67.2 33.5 4.3 14.7 67.3
SeOB1SBO1HZO1HSBO (6 : 4 : 5 : 5) 32.7 5.5 14.2 68.1 32.5 5.6 14.1 68.2
SeOB1SBO1CSO1HSBO (6 : 4 : 5 : 5) 33.0 5.6 14.3 67.7 32.6 5.6 14.1 68.1
SeOB1SBO1SFO1HROO (6 : 4 : 5 : 5) 33.0 4.4 14.6 67.5 33.6 4.2 14.5 67.1
2
R 0.989 0.984 0.996 0.992

§ Abbreviations of oils/oil mixtures are defined in the legend of Fig. 1.

Tab. 2. Effect of number and combination of the wavelengths used for training of the ANN on the estimation capability of
the network for the test samples.
2
Number Number of Wavelength combination for Average of correlation coefficients (R )
2
of input wavelength which the highest R value was for test samples
variables combinations determined [nm] §
L* a* b* Average
6!
1 6!ð6 6Þ! ¼ 1 416, 456, 483, 537, 611 and 672 0.993 0.990 0.996 0.993
6!
2 5!ð6 5Þ! ¼ 6 456, 483, 537, 611 and 672 0.993 0.988 0.993 0.991
6!
3 4!ð6 4Þ! ¼ 15 456, 537, 611 and 672 0.994 0.988 0.992 0.991
6!
4 3!ð6 3Þ! ¼ 20 483, 537 and 611 0.994 0.989 0.992 0.992
6!
5 2!ð6 2Þ! ¼ 15 483 and 537 0.993 0.980 0.992 0.988
6!
6 1!ð6 1Þ! ¼ 6 537 0.976 0.959 0.986 0.974
§
Average of correlation coefficients of L*a*b* used for the selection of the best wavelength combinations.

 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim

www.ejlst.com
Eur. J. Lipid Sci. Technol. 109 (2007) 157–164A new method for color determination of edible oils in L*a*b* format 163

wavelengths used for training of the ANN on the estima-tion 4 Conclusions

capability of the network for the test samples is given in
Tab. 2. Although the ANN trained with one wavelength had Edible oils have colors ranging from light yellow to dark
2
a good average R (0.974) for L*a*b* and could suc- green or brown, depending on the type and concentration of
cessfully estimate L*a*b*, increasing the number of the color pigments. The concentration of these pig-ments
wavelengths also increases the correlation coefficients for not only depends on the type of plant, but also on the type
2 of process. Therefore, color determination is an
the test samples, and the best average R (0.993) was
indispensable kind of analysis for the edible oil industry
obtained when the ANN was trained with the data of six
before marketing. Currently, the available official meth-ods,
wavelengths. The wavelength of 537 nm was included in all
except the spectrophotometric one, require visual judgment
six wavelength combinations, and 483 nm was inclu-ded in
and/or more oil samples. This study shows that ANN can be
the first five wavelength combinations (Tab. 2). The
absorption peak at 483 nm may be attributed to b-caro-tene used effectively to determine the color of most edible oils, a
because Melendez-Martinez et al. [10] have reported that b- sesame oil blend, and of heated oils as well. Further
carotene has absorption peaks at around 450 and 480 nm. research needs to be conducted to examine the
It is apparent that the absorbance peak at 537 nm strongly performance of ANN for the color determination of liquid
affects the L*a*b*DE values, and the color of the oils as food samples, such as clear fruit juice, and also non-food
well. Since Merzlyak et al. [21] have reported that samples, such as inks which are non-transpar-ent and more
anthocyanins have absorption maxima near 550 nm, the viscous than edible oils and widely used in the textile or
absorption peak detected at 537 nm may be attribut-ed to paint industries.
these flavonoids.

Instrumental measurement of oil color is a very useful tool References

for quality control of oil in the edible oil industry because it is
an objective and rapid technique that enables the ana-lyst to [1] R. D. O’Brien: Fats and Oils. Formulating and Processing
for Application. Technomic Publishing Company,
obtain L*a*b* values in a few seconds. Besides having all
Lancaster, PA (USA) 1998.
the advantages of an instrumental color meas-urement, the [2] S. Fengxia, Z. Dishun, Z. Zhanming: Determination of oil
method developed in this study is cost effective because it color by image analysis. J Am Oil Chem Soc. 2001, 78,
uses a simple UV/visible spectro-photometer readily 749– 752.
available in many laboratories and disposable cuvettes [3] AOCS: Official Methods and Recommended Practices of the
American Oil Chemists’ Society. Section C: Commercial
allowing easy color measurement. The method makes it th
Fats and Oils. 5 Edn. Ed. D. Firestone, AOCS Press,
possible to avoid operator variability associated with visual Champaign, IL (USA) 1999.
comparison. Because of using a large number of training [4] P. Joshi: Color measurement of foods by color reflectance.
and validation data sets, results from the ANN agree closely In: Color in Food: Improving Quality. Ed. D. MacDougall,
CRC Press, Washington DC (USA) 2002, pp 21–30.
with those obtained with the Minolta spectrophotometer over
[5] K. L. Yam, S. E. Papadakis: A simple digital imaging
a wide color range of oil samples. The oils to be measured method for measuring and analyzing color of food surfaces.
in color with the devel-oped method should be completely J Food Eng. 2004, 61, 137–142.
liquid and clear, otherwise the impurities in the oil will affect [6] K. Leon, D. Mery, F. Pedreschi, J. Leon: Color
the absorb-ance and also the L*a*b*DE values. The oil measurement in L*a*b* units from RGB digital images.
Food Res Int. 2006, 39, 1084–1091.
samples used in this study have wide ranges of a* and b*
[7] B. Onal, G. Ergin: Antioxidative effects of a-tocopherol and
values, but close L* values. Therefore, the ANN should be ascorbyl palmitate on thermal oxidation of canola oil. Nah-
recali-brated for oils having L* values much lower or higher rung/Food. 2002, 46, 420–426.
than the ones used in this study. [8] J. Lu, J. Tan, P. Shatadal, D. E. Gerrard: Evaluation of pork
color by using computer vision. Meat Sci. 2000, 56, 57–60.
[9] M. G. O’Sullivan, D. V. Byrne, H. Martens, L. H.
Gidskehaug, H. J. Anderson, M. Martens: Evaluation of
Refined edible oils do not contain fluorescent compounds pork color: Pre-diction of visual sensory quality of meat
such as mycotoxins, because most of such compounds, from instrumental and computer vision methods of color
analysis. Meat Sci. 2003, 65, 909–918.
extracted from oil seeds during oil extraction with hexane,
[10] A. J. Melendez-Martinez, G. Britton, I. M. Vicario, F. J. Her-
are removed during deodorization [23]. However, myco- edia: Relationship between the color and the chemical
toxins can be easily extracted into the oil phase during structure of carotenoid pigments. Food Chem. 2007, 101,
pressing or centrifugation of the oil from olive fruits; so they 1145–1150.
might be found in olive oils that are not subjected to refining [11] M. Garcia-Esteban, D. Ansorena, O. Gimeno, I. Astiasaran:
Optimization of instrumental color analysis in dry-cured
processes. Therefore, the developed method should be ham. Meat Sci. 2003, 63, 287–292.
calibrated in the presence of fluorescent com-pounds if the [12] T. Brosnan, D. W. Sun: Improving quality inspection of food
color of non-refined oils is to be measured. products by computer vision. J Food Eng. 2004, 61, 3–16.

 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ejlst.com

164 K. Kılıc¸ et al. Eur. J. Lipid Sci. Technol. 109 (2007) 157–164

[13] D. W. Hatcher, S. J. Symons, U. Manivannan: olive fruit on the chlorophyll and carotenoid fractions of the
Developments in the use of image analysis for the drupes and virgin olive oils from Arbequina and Farga culti-
assessment of oriental noodle appearance and color. J vars. Food Chem. 2007, 100, 748–755.
Food Eng. 2004, 61, 109– 117.
[14] W. Lou, S. Nakai: Application of artificial neural networks for [20] A. N. Vinha, F. Ferreres, B. M. Silva, P. Valentio, A. Gon-
çalves, J. A. Pereira, M. B. Oliveira, R. M. Seabra, P. B.
predicting the thermal inactivation of bacteria: A combined
Andrade: Phenolic profiles of Portuguese olive fruits (Olea
effect of temperature, pH and water activity. Food Res Int.
europaea L.): Influences of cultivar and geographical
2001, 34, 573–579.
origin. Food Chem. 2005, 89, 561–568.
[15] C. J. Du, D. W. Sun: Recent developments in the applica-
tions of image processing techniques for food quality eval- [21] A. N. Merzlyak, A. E. Solovchenko, A. A. Gitelson: Reflec-
uation. Trends Food Sci Technol. 2004, 15, 230–249. tance spectral features and non-destructive estimation of
˙ chlorophyll, carotenoid and anthocyanin content in apple
[16] D. Bas¸ , F. C. Dudak, I . H. Boyacı: Modeling and optimization.
III. Reaction rate estimation using artificial neural network fruit. Post Harvest Biol Chem. 2003, 27, 197–211.
(ANN) without a kinetic model. J Food Eng. 2007, 79, 622–
628. [22] S. Paul, G. S. Mittal: Dynamics of fat/oil degradation during
˙ frying based on optical properties. J Food Eng. 1996, 30,
[17] K. Kılıç, I . H. Boyacı, H. Köksel, I. Küsmenog˘lu: A classifica- 389–403.
tion system for beans using computer vision system and
artificial neural networks. J Food Eng. 2007, 78, 897–904. [23] S¸ . Aktug˘ Gönül: Dig˘er gıdalarda mikrobiyolojik bozulmalar,
[18] F. Pedreschi, J. Leon, D. Mery, P. Moyano: Development patojen mikrorganizmalar ve muhafaza yöntemleri. In:
of a computer vision system to measure the color of potato Gıda Mikrobiyolojisi. Eds. A. Ünlütürk, F. Turantas¸ , Mengi
chips. Food Res Int. 2006, 39, 1092–1098. ˙
Tan Basımevi, I zmir (Turkey) 1999, pp. 403–423.
[19] M. N. Criado, M. J. Motilva, M. Goni, M. P. Romero: Com-
parative study of the effect of the maturation process of the [Received: September 29, 2006; accepted: December 5, 2006]

 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.ejlst.com