Sensors and Actuators A 109 (2004) 195–201

Interference-enhanced imaging for detectingoil layer floating on the water

N. Saitoa,∗, K. Takizawab, T. Kurokawac

a NHK Science and Technical Research Laboratories, Setagaya, Tokyo 157-8510, Japanb Seikei University, Musashino, Tokyo 180-8633, Japan

c Tokyo University of Agriculture and Technology, Koganei, Tokyo 184-8588, Japan

Abstract

High-contrast imaging of thin oil layer on the water has been proposed as subtraction of two images taken through different band-passfilters. There is brightness difference in the part of the oil between two pictures taken using two band-pass filters with different centerwavelengths because of the interference effects. On the other hand, there is not large brightness difference in the part of the water withoutthe oil. Subtraction of one picture from the other can make the part of the oil brighter than that of the surface of the water. Calculation ofthe brightness and primitive experiments of the imaging have shown that enhancement in the contrast is obtained and is useful for the oillayer detection on the water.© 2003 Elsevier B.V. All rights reserved.

Keywords: Interference; TV camera; Band-pass filter; High-contrast imaging; Oil film

1. Introduction

When a ship or an airplane meets with an accident inthe ocean, its oil spill is used as a clue to search it. Ifhigh-contrast images between the oil layer floating on thesea and the surrounding sea are available, quick detectionmight be possible. High-contrast imaging of the oil filmsis important and useful not only for the search, but also forthe television report of, for example, an oil belt on the sea.Interference may help to improve the contrast ratio, becausethe reflection spectrum of the oil film on the water showsstrong wavelength dependence, while reflection from thewater surface without the oil does not so strongly dependon the wavelength. High-contrast imaging of the oil filmon the water will be realized owing to this difference. Thehigher the contrast is, the more quickly we can expect tofind the victim. Moreover, if the regions with and withoutthe oil films are clearly distinguished through high-contrastimages, the oil films can be detected without a need of hu-man watches. This automatic detection will lead to quickerand more accurate detection of oil films.

In this paper, attempts have been made through calcula-tion and experiments to confirm the possibility of contrastenhancement by this method. Since a set of band-pass filters

∗ Corresponding author.E-mail address: saito.n-fk@nhk.or.jp (N. Saito).

are characterized by center wavelengths, bandwidth, andpeak transmittance, and reflection spectrum depends on thethickness of the oil film, incident angle, and polarization,we have investigated effects of these parameters on theenhancement. Preliminary experiment supported the resultsobtained from the calculation.

2. Principle and calculation

The enhancement of the contrast ratio is based on the prin-ciple that the reflection from the oil film on the water stronglydepends on wavelength owing to the interference caused bythe multiple reflections at the top and bottom interfaces ofthe oil, whereas, when there is no oil on the surface, the re-flection does not depend on the wavelength so strongly. Thisdifference will lead to high-contrast ratio detection of the oilfilm on the water. Consider a set of optical band-pass filtersare used to observe the water surface with and without oilfilm. If one filter is tuned to the maximum of the reflectionspectrum of the oil film and the other is set to the minimum,the absolute value of the subtraction of one spectrum fromthe other will give a large value for the oil film, whereas asmall value (nearly null) for the water without the oil. Thismeans that we can obtain a large contrast between the tworegions with and without the oil through the present method.Moreover, we can easily get a clear image of the oil film bydigitizing the picture, because we can set the brighter region

0924-4247/$ – see front matter © 2003 Elsevier B.V. All rights reserved.doi:10.1016/j.sna.2003.09.012

196 N. Saito et al. / Sensors and Actuators A 109 (2004) 195–201

camera

Vat Water

White paper

Image processing unit

Wave Form Monitor Display

VideoCassette Recorder

Optical bandpass filters

Light source

Heavy Oil film

Fig. 1. Experimental setup for measurement and signal processing of reflectance of water with and without oil film.

(the oil films) as “1” and the darker region (the water withoutoil) as “0”. This method can easily expand to the automaticdetection of oil on the water. Although the principle is com-mon, the idea is useful to provide us with quick and accuratemeans of detection of the oil films floating on the sea.

The proposed setup for the enhanced detection of oil filmis schematically shown inFig. 1. In this setup, one cameraand two band-pass filters are used. One filter is placed infront of the camera. After one picture is stored in a framememory, the filter is replaced by the other band-pass filterthat has different optical characteristics from the first filter.Then the picture is taken with the second filter. The absolutedifference is calculated on pixel-by-pixel basis.

In order to simulate the above circumstances, theoreticalanalysis of the interference behavior was performed accord-ing to the following procedure:

1. Reflection spectrum was obtained for a three-layermedium consisting of air, oil, and water, and for atwo-layer medium (air and water).

2. Intensity of light that reaches a photo-detector after pass-ing through a band-pass filter was calculated for the re-flected light from the two types of the media.

3. Set of parameters was calculated for two band-pass filtersthat give maximum contrast ratio.

Reflectance was calculated for the three-layer medium, air(medium 1, refractive indexn = n1), oil film (medium 2,uniform thicknessd, n = n2), and water (medium 3,n =n3), for the light (wavelengthλ and incident angleφ1) ac-cording to the formula in references[1,2]. Transmittanceand reflectance at the interface of the mediam andn (m, n= 1, 2, 3) istmn andrmn, respectively. The total reflectanceR is obtained as the infinite summation of reflectances at theinterfaces,

Rx(λ) =r12x

2 + {r23x exp(−αL)}2

− 2r12xr23x exp(−αL)cosΨ

1 + {r12xr23x exp(−αL)}2

− 2r12xr23x exp(−αL)cosΨ

(1)

where suffixx denotesp or s polarization,α the absorptioncoefficient,L the thickness of the oil layer, andΨ is thephase change in the medium 2. When there is no oil on thewater, the reflectance is given by a conventional equation ofair and water.

In this paper, calculation was done fors- and p-polarization. Final result is obtained as the average of thetwo. Ther23, r21, t21, andt12 are obtained from Snell’s lawusing n1, n2, n3, andφ1. These parameters were obtainedas follows. The refractive index of the air,n1, was set to beunity. The measured values were used for the medium 2,heavy oil A (JIS K2205–1980, TYPE 1). Its refractive in-dex,n2, was measured with a refractometer (CARL ZEISSJENA) (Fig. 2a), while its absorption coefficient,α, was

Fig. 2. Spectra of (a) refractive index and (b) absorption coefficient ofthe heavy oil A.

N. Saito et al. / Sensors and Actuators A 109 (2004) 195–201 197

Fig. 3. Reflection spectra of water with and without oil film, and transmission spectra of two band-pass filters.

obtained using a spectrophotometer (HITACHI) (Fig. 2b).The refractive index of the water,n3, was taken from[3]. In the calculation, the least-square-fitted values wereused.Since wavelength dependence of the reflected light isobtained asR(λ) × S(λ), where S(λ) is the spectrum ofthe light source, the intensity of the light that goes throughan optical band-pass filter, whose transmittance isT(λ), isobtained as an integral ofR(λ) × S(λ) × T(λ)

P =∫ λm+W/2

λm−W/2R(λ)S(λ)T(λ) dλ (2)

whereλm andW are the center wavelength and bandwidthof the band-pass filter, respectively. In the present calcula-tion, P was obtained as the summation of values calculatedat 1 nm step. We setS as constant unity for the simplicity ofthe calculation. We setT as unity in the wavelength region�m−W/2 < λ < λm+W/2 and as null whenλ < λm−W/2or λ > λm + W/2 also for the simplicity of the calculation.Absolute differences were calculated between the signalintensity with one filter and that with the other.

Since what determines the quality of the image is theratio of signal to background, we used enhancement factorC in this paper for the evaluation, which was defined asthe ratio of the absolute difference of theP for the oil film(signal) to that of theP for the surface of the water withoutoil (background).

C=∣∣∣∣ A − B

A′ − B′

∣∣∣∣ (3)

whereA andB are maximum and minimum of reflectancefrom the water with an oil film, respectively, andA′ andB′are those of reflectance from the water without oil (Fig. 3).

3. Experimental

Experimental setup is schematically shown inFig. 1.Heavy oil A (JIS K2205–1980, TYPE 1) was spilt on the

water in a vat using a pipette. A sheet of white paper wasused to diffuse the light from the light source, and to blockunnecessary reflection. In order to avoid the reflection fromthe bottom of the vat, black ink of about 1 % wasadded tothe water to make it opaque.

Images were taken using an image-processing unit(GDIPUN-C3, SHARP SEMICONDUCTOR) as follows:The frame memory in the unit stores a picture taken withone optical band-pass filter set in front of the CCD camera.After the filter is replaced by the other one, the second pic-ture is stored in the frame memory. Then, the unit calculatesthe absolute difference of the two pictures on pixel-by-pixelbasis. After the image-processing unit digitizes the aboveresult at a threshold level, the final result is displayed on themonitor. In the present experiment, the threshold level wascalculated as the average brightness over the entire image ofthe absolute difference of the two pictures. In the digitizingprocess, the pixels brighter than the threshold level were set“1”, while those darker than that were set “0”.

Several pairs of band-pass filters were prepared accord-ing to the result of the calculation. The enhancement wasquantitatively expressed using a videocassette recorder anda waveform monitor. Since emission of the real light sourcehas wavelength dependence and the transmittance are differ-ent from one filter to another for the real filters, iris of thecamera was adjusted, before the second picture is stored, toregulate the incident light so that the brightness of the partof the water surface becomes equal for the two pictures.

4. Results and discussion

4.1. Results of the calculation

In Fig. 4, calculated signal intensity spectra are shownfor the oil films with various thicknesses (0.1, 0.3, 1, and3�m) using filters of three kinds of bandwidths (1, 10,and 100 nm). In the calculation, the incident angle was set

198 N. Saito et al. / Sensors and Actuators A 109 (2004) 195–201

Fig. 4. Calculated spectra of signal intensity for the oil films with various thicknesses 0.1–3�m, and various bandwidths 1, 10, and 100 nm; solid lines:reflection from the oil film on the water; broken lines: reflection from the water surface.

constant 40◦. When the oil layer is thin (0.1 and 0.3�m), theoscillation in the spectra is almost the same as the reflectionspectra, even with the band-pass filter of 100 nm. On theother hand, when the oil layer is thick (1 and 3�m), with the

1 10 1000

100

200

300

400

bandwidth [nm]

Enh

ance

men

t fac

tor

1 10 1000

10

20

30

bandwidth [nm]

Enh

ance

men

t fac

tor

1 10 1000

100

200

bandwidth [nm]

Enh

ance

men

t fac

tor

1 10 1000

20406080

100

bandwidth [nm]

Enh

ance

men

t fac

tor

1 10 1000

100

200

bandwidth [nm]

Enh

ance

men

t fac

tor

(a) 0.1 µm (b) 0.3 µm

(c) 0.7 µm (d) 1 µm

(e) 3 µm

Fig. 5. Calculated relation between bandwidth and enhancement factor as a parameter of the thickness of the oil layer.

bandwidth is 100 nm, amplitude in the oscillation is smallerthan that with the bandwidth of 1 and 10 nm. This is becausethe distance between the peak and the valley becomes equalto or smaller than the bandwidth, when the oil layer is thick.

N. Saito et al. / Sensors and Actuators A 109 (2004) 195–201 199

0 1 2 3 410

50

100

500

Thickness of oil film [ m]

Opt

imum

ban

dwid

th [n

m]

µ

Fig. 6. Oil thickness dependence of optimum bandwidth obtained fromFig. 5.

Fig. 5 shows the calculated result, obtained fromFig. 4,of relation between bandwidth and enhancement factorCas a parameter of the thickness of the oil layer. FromFig. 5,optimum bandwidth was determined as the wavelength re-gion that gives 95% of the maximum enhancement factorFig. (6). Since not only enhancement of the image but alsothe quantity of light must be considered to achieve largesignal-to-noise ratio, bandwidth as large as possible wasnecessary. We can predict fromFig. 6 that preferable band-width of the filters is about 50 nm for the oil films withabout 1�m thickness.

Fig. 7. Experimental images obtained using filters of bandwidth 40 nm and center wavelengthsλ1, λ2 = 500 and 650 nm.

4.2. Results of the experiments

Band-pass filters of which bandwidth is 40 nm were pre-pared according to the results ofSection 4.1. To confirm theeffect of the optimum filters, 10 nm-bandwidth filters werealso prepared. Experiments were conducted using these fil-ters. Fig. 7 shows a typical result obtained with filters of40 nm bandwidth and center wavelength of 500 nm (λ1) and650 nm (λ2). Fig. 7ashows the image taken without usingband-pass filters.Fig. 7b and 7care the results taken with theband-pass filters ofλ1 andλ2. The digitized picture of theabsolute difference image is shown inFig. 7dIn the digitiz-ing process, the threshold level was calculated as the averagebrightness over the entire image of the absolute differenceof the two pictures (b) and (c), and the pixels brighter thanthe threshold level were set “1”, while those darker than thatwere set “0”.Fig. 8shows the results taken with the filters of(a) 10 nm bandwidth andλ1, λ2 = 550, 650 nm, (b) 10 nmbandwidth andλ1, λ2 = 600, 650 nm, (c) 40 nm bandwidthand λ1, λ2 = 550, 650 nm, and (d) 40 nm bandwidth andλ1, λ2 = 600, 650 nm. The inhomogeneity in the intensityof the reflected light observed in the pictures indicates thatthe thickness of the oil layer on the water was not uniform.

Images taken with 40 nm bandwidth filters were clearerthan those with 10 nm bandwidth filters. Quantitativecomparison was made as the maximum TV signal level

200 N. Saito et al. / Sensors and Actuators A 109 (2004) 195–201

Fig. 8. Digitized images obtained from the absolute difference of two pictures using various combinations of band-pass filters: (a) 10 nm-bandwidthfilters of λ1, λ2 = 550, 650 nm; (b) 10 nm-bandwidth filters ofλ1, λ2 = 600, 650 nm; (c) 40 nm-bandwidth filters ofλ1, λ2 = 550, 650 nm; and (d)40 nm-bandwidth filters ofλ1, λ2 = 600, 650 nm.

Table 1Experimental and calculated results of image signal level of the absolute subtracted images

Bandwidth(nm)

Centerwavelengths(nm)

Experimental Calculated maximumrelative value

Average thicknessof oil film (�m)

Maximum image signallevel of oil layer

Image signal levelof water surface

(c)/(d) Relativevalue

(a) (b) (c) (d) (e) (f) (g) (h)

10 550, 650 8 3.6 2.2 1 1 0.80600, 650 10 5.0 2.0 0.91 0.33 0.20

40 500, 650 28 4.6 6.1 2.77 3.72 0.36550, 650 19 3.4 5.6 2.55 1.54 0.22600, 650 30 5.6 5.4 2.45 1.28 0.31

of the oil film and the water surface (Table 1). Values inthe column (d) must be almost null, if our calculation canbe applied to the real oil layers. The difference betweenthe calculation and the experiment is probably caused byinsufficient adjustment of the iris of the camera.

4.3. Discussion

Comparison of the results of calculation and experimentis made inTable 1. Since we cannot obtain experimentalvalue of the enhancement factor, relative values are shown

to the signal level of the combination of 550 and 650 nmof 10 nm bandwidth. InTable 1, thickness of oil film wasinferred as (volume of the oil drop)/(area of the oil film).

The order of the experimental relative value (column (f))is the same as that of the calculated one (column (g)). Thismeans that the present method is effective in obtaining en-hanced images.

In the present setup, however, no movement of subjectsand observer was assumed. Therefore, adjustment in thebrightness was done to take the second image after one im-age was taken. However, oil film on the water sometimes

N. Saito et al. / Sensors and Actuators A 109 (2004) 195–201 201

became broader. This might be the difference between thecalculation and the experiment. Reduction in time lag be-tween two pictures will lead to better agreement in the cal-culation and the experiment.

In the calculation, we just determined two center wave-lengths, which give maximum and minimum reflectance fora certain value of the thickness, and neglected the effect ofthe film thickness on the center wavelength and bandwidthof the filter. However, we do not know the exact value of thethickness of the actual oil films. If the thickness of an oil filmcoincides with the thicknesses that give the maximum andthe minimum for the two filters, we can achieve maximumcontrast, as calculated. On the other hand, if they hit uponthe values that give the same reflectance for the two filters,the contrast is null. Moreover, as the experiment showed,oil film is not uniform in thickness but changes from placeto place. It means that the contrast changes depending onthe position. However, since contrast is almost null for allthe surface of water without oil film, we can distinguish theregion with the oil film from that without the oil, and theresult of the present method remains unchanged.

Although dilute (∼1%) black-ink water was used in thepresent experiment, the effect of the ink to the results is notso large, because the influence of the ink to the refractiveindex of the water is expected to be small. Even if the changeis so large that the wavelengths that gives the maximumand the minimum of the reflectance change, our conclusionremains unchanged that we can distinguish the part with anoil film from that without one.

Furthermore, although, in the present analysis, waves andwind were neglected, their effects must be critical and cannotbe neglected. Since their effects will be analyzed in anotherarticle precisely, we discuss them qualitatively. Their effectscan be taken into account by considering the incident anglechanges (1) every moment (depending on time) and (2) fromplace to place (depending on position), while it was regardedas constant in the present analysis.

1. When the incident angle changes depending on time, theproposed method, which uses two pictures taken at dif-ferent times, can do nothing. However, we can solve theproblem by an idea, in which1.1. one picture is split into two identical pictures;

1.2. two pictures are taken by different cameras;1.3. two filters are designed according to the present me-

thod, and each filter is placed in front of the camera.The present method can be applied to the difference ofthe two pictures that are taken at the same instance.

2. We can deal with position-dependent incident angle caseby the same technique as the case in which the thicknessof oil film changes from place to place. Since the positionwhere contrast takes a maximum value and that where ittakes a minimum one appear alternately, a stripe patternis seen on the oil film. When both the incident angle andthe thickness change, the stripes bend depending on thedirection of the change. Therefore, the existence of theoil film can be detected, in these cases, and the mainresults remains unchanged.

5. Conclusion

Enhancement in the contrast ratio of the image of oilfilm is proposed by means of the subtraction of two imagestaken with two optical band-pass filters with different cen-ter wavelengths. Calculation has been done to confirm thatthis method is effective. Qualitative agreement was obtainedbetween the results of calculation and primitive experiment.Better agreement will be obtained if time lag for taking asecond picture is reduced.

Acknowledgements

The authors thank S. Saito for his assistance in thecalculation and experiments. They also acknowledge N.Kawamura and H. Fujikake for their useful discussions.

References

[1] M. Born, E. Wolf, Principles of Optics, sixth ed., Pergamon Press,Oxford, 1980.

[2] A.W. Crook, J. Opt. Soc. Am. 38 (1948) 954.[3] V.M. Zolotarov, V.N. Morozov, E.V. Smirnova, Optical Constants of

Natural and Artificial Media, Khimiya, Leningrad, 1984, p. 14.