Instrument to measure the bidirectional reflectance
- 1. Instrument to measure the bidirectional reflectance distribution function of surfaces Kenneth J. Voss, Albert Chapin, Marco Monti, and Hao Zhang A new instrument to measure the in situ bidirectional reflectance distribution function ͑BRDF͒ of surfaces is described. This instrument measures the BRDF for eight illumination angles from 0 to 65 deg, three colors ͑475, 570, and 658 nm͒, and at over 100 selected viewing angles. The viewing zenith angles range from 5 to 65 deg, and the azimuth angles, relative to the illumination direction, range from 0 to Ϯ180 deg. Many tests of the system have been run and show that for flat surfaces the BRDF of a sample surface can be measured with a precision of 1–5% and an accuracy of 10% of the measured reflectance. The BRDF for a dry and wet sand sample is presented as a demonstration of the instru- ment. © 2000 Optical Society of America OCIS codes: 120.5700, 120.4640, 240.6700, 010.4450. 1. Introduction bertian reflector, illuminated uniformly, the radiance The bidirectional reflectance distribution function reflected from the surface is independent of the view ͑BRDF͒ is an important parameter to describe the angle and illumination direction. In this paper the interaction of light with a surface. The BRDF de- data are presented as the reflectance factor REFF͑i , scribes the variation of reflectance with the view and v, ͒. The REFF͑i , v, ͒ is defined as the ratio of illumination direction. To define the BRDF we first the reflectance of a surface to that of a perfect Lam- need to define two other terms in radiometry. The bertian surface with the same illumination and view radiance L͑v, ͒ is the power per unit solid angle conditions. The REFF͑i , v, ͒ is related to the centered in a given direction ͑v, ͒ passing through BRDF͑i , v, ͒ by a simple factor of ͓i.e., REFF͑i , a unit area perpendicular to the direction ͑units are v, ͒ ϭ BRDF͑i , v, ͔͒. We use this factor in- in W cmϪ2 srϪ1͒. The irradiance is the power pass- stead of the BRDF because it is easy to quickly see the ing through a unit area, with collimated irradiance difference between the measured surface and a Lam- being the power in a collimated beam passing bertian reflector. Although many surfaces are as- through an area perpendicular to the direction of the sumed to be Lambertian, the reality is that few beam ͑units are in W cmϪ2͒. surfaces are. In general, even for surfaces that are The BRDF is defined as the ratio of the radiance close to Lambertian, the assumption will fail in two L͑v, ͒ scattered by a surface into the direction ͑v, areas. The first obvious failure is that many flat ͒ ͑see Fig. 1 for a coordinate description͒ to the surfaces have an enhanced specular component, i.e., collimated irradiance E incident on a unit area of the an increase in the reflectance when the illumination surface1: and viewing polar angles ͑i and v, respectively͒ are the same, but the relative azimuth ͑͒ is 180 deg ͑the BRDF͑i , v, ͒ ϭ L͑v, ͓͒͞E cos͑i ͔͒. direction a mirror would send light͒. Many rough surfaces also show a reflectance enhancement in the A common simplification of the BRDF is to assume hot-spot direction, when i ϭ v and ϭ 0 deg ͑view- a Lambertian reflector.2 With an infinite ideal Lam- ing along the illumination beam͒. This hot spot has been seen from many natural surfaces and is a com- bination of many factors including a lack of shadows The authors are with the Department of Physics, University of along this direction.3 Miami, Coral Gables, Florida 33124. The e-mail address for K. J. Voss is voss@physics.miami.edu. To model the light field near a surface, the BRDF of Received 27 September 1999; revised manuscript received 9 Au- the surface must be known. As part of the U.S. Of- gust 2000. fice of Naval Research Coastal Benthic Optical Prop- 0003-6935͞00͞336197-10$15.00͞0 erties Program, enhanced models of the light field © 2000 Optical Society of America near benthic surfaces in the water are being devel- 20 November 2000 ͞ Vol. 39, No. 33 ͞ APPLIED OPTICS 6197
- 2. Fig. 1. Diagram of measurement geometry for BRDF. Colli- mated irradiance E comes in at angle i . The reflected radiance is Fig. 2. Diagram of instrument design with major components sampled at a polar angle v with azimuth . listed. oped. Thus there was a requirement to determine flected light from the surface is collected by the view accurately the BRDF for natural surfaces, such as fibers, which carry this reflected light to an array. sand. Each view fiber comes into the array at a different Although the BRDF is an important factor in de- spot, and this array is imaged on a charge-coupled scribing the surface reflectance, it can be difficult to device ͑CCD͒ camera. The array is displayed in the measure, even for simple surfaces. Because the camera image as an array of bright dots; the bright- function varies with both illumination angle and ness of each dot represents the reflectance into the viewing angle, many measurements are required to specific angle. The instrument housing also con- properly characterize a surface. Typically these tains the control computer and a hard drive for data measurements have been performed by an instru- storage. The whole system can be powered by a bat- ment with which individual measurements of view- tery, allowing remote operation. The instrument ing directions are performed for each illumination case is a 39-cm-high cylinder with a diameter of 42 condition.4,5 The viewing directions and illumina- cm. The specific instrument components are de- tion conditions can be varied either manually or au- scribed below. tomatically. In either case there is a large number of individual measurements to be performed. This A. Light Source is time-consuming, few measurements have been per- To measure reflectance accurately one must have sta- formed, and these have been laboratory measure- ble, even illumination of the surface and view this ments for small surfaces. To make these surface with a stable detector. There are two measurements in situ for benthic surfaces requires a choices for the basic optical design of a reflectometer. different design because the instrument would be op- In one of these, the surface must be illuminated over erated by a scuba diver with limited air and measure- a large area, and the instrument must view a portion ment time. of the illuminated surface. This choice allows the We developed an instrument design that allows instrument to be relatively insensitive to the distance many viewing angles to be measured simultaneously between the receiver and the surface rrs ͑ignoring for a specific illumination condition. With multiple attenuation in the path͒. This is because the change illumination wavelengths and angles, the spectral in the observed area of the surface that is due to a BRDF can be determined in a time suitable for mea- change in rrs is compensated by a change in the re- surement in situ. This design can also be extended ceived flux that is due to a 1͞rrs2. However, this into the laboratory for quick, accurate characteriza- measurement geometry requires a well-defined re- tion of both natural and man-made surfaces. ceiver aperture and wastes some of the illuminating beam. The second choice is to have a well-defined 2. Instrument Description small illumination area and overview the surface. A schematic of the instrument design is shown in Fig. In this case the constraints on the receiver optics can 2. The two most important features of the instru- be relaxed as it is required that only a large area, ment are that multiple viewing angles are measured including all the illuminated area, be viewed. How- simultaneously and there are no moving parts in the ever, this choice is sensitive to rrs because when rrs is design. Illumination is provided by three colors of increased, the increased area viewed by the receiver light-emitting diodes ͑LED’s͒ which illuminate indi- is not illuminated. Even with this consideration, vidual fibers. These fibers guide the light to a ball our design uses this geometry because we have many lens which collimates the light and directs it onto the receivers ͑of the order of 100͒, and it is more efficient sample surface. We select the color and illumina- to control the illumination optics than the receiver tion angle by turning on individual LED’s. Re- optics. 6198 APPLIED OPTICS ͞ Vol. 39, No. 33 ͞ 20 November 2000
- 3. Illumination is provided by blue, green, and red were drilled in the dome pointed at the center of the LED’s ͓Panasonic LNG992CFBW, Lumex SSL- sample area. These holes were drilled at the polar LX1001333XGC, and Panasonic LN261CAL͑UR͒, re- angles of 5°, 15°, 25°, 35°, 45°, 55°, and 65°. They spectively͔. There are separate LED’s for each were also dispersed azimuthally, with a concentra- illumination angle ͑i ͒ from 0° to 65° ͑0°, 5°, 15°, 25°, tion of measurement locations in the forward ͑spec- 35°, 45°, 55°, 65°͒. A 1-mm plastic fiber is used to ular͒ and backward ͑hot-spot͒ directions as this was carry the light from the illumination block to a ball anticipated to be an important feature of many sur- lens which produces the collimated light that illumi- faces. Hence azimuthal angles ͑͒ measured are nates the sample surface. We wanted the illumi- Ϯ5°, Ϯ10°, Ϯ15°, Ϯ30°, Ϯ45°, Ϯ60°, Ϯ75°, Ϯ90°, nated area to be the same for each LED color, so we Ϯ105°, Ϯ120°, Ϯ135°, Ϯ150°, Ϯ165°, Ϯ172°, and could not have a separate fiber for each color at a Ϯ180°. At the smaller polar angles some of the az- specific i because multiple fibers at the focal plane of imuthal angles are missing because they physically the ball lens would be directed into different direc- interfere with the illumination optics. All the avail- tions. To get the light from three LED colors into able spaces on this grid were drilled, and choices were the illumination fiber, a cavity was made in the illu- made about which 100 viewing directions would be mination block to allow the three LED’s, for a specific used. We concentrated on the forward and the back- i , to point at the tip of the fiber. The cavity was made from polished aluminum; and, although the ward directions and spreading the available fibers in fiber was not illuminated as efficiently as it would be the plus and minus azimuthal directions so that the for a single LED, it does allow a single fiber to be used sample symmetry could be determined. For many for each illumination angle. natural, in situ samples this symmetry could not be The light from the fiber is collimated with 10-mm assumed. BK-7 ball lenses ͑Edmund Scientific͒. The fibers are Illumination and view fibers are limited to polar arranged so that the center of the illuminated spot on angles Ͻ65° because of mechanical constraints. Be- the sample surface remains constant, independent of cause we were concerned with natural surfaces that illumination angle. There is a small change of the have randomly oriented facets, if any facets at all, we illuminated surface that is due to the circular spot expected the largest peaks to be in the hot-spot or becoming oval with increased illumination angle. specular direction. Thus we had view fibers at the The light from the ball lens exits through a hemi- same polar angles as the illumination fibers, and we spherical window over the sample area. We used a tried to concentrate the measurements in the forward 20-cm-diameter compass dome ͑Rule Industries͒ as and the backward directions. Because we have dis- the system window. The size of this hemisphere in- crete sampling directions we may miss sharp reflec- volves several competing factors. The larger the tion features. But two factors make this less dome, the less likely that the measurement will be important. First, with the natural samples it is less affected by multiple reflections. Also, large sizes are likely that we will have sharp reflection features preferred as deviations of the sample height are less other than in the specular and hot-spot directions. critical because these affect the measurement by the Second, the data from this instrument will be used inverse square of the relative change in distance. predominantly in models with less angular resolution Unfortunately the ultimate instrument size is criti- than our measurements. Thus we do not consider cally dependent on the size of this dome, and for that our resolution limits the applicability of our data diver-operated instruments, the smaller the better. significantly. Also, large dome sizes require longer path lengths The fibers have a numerical aperture of 0.58. The through the medium, hence the opportunity for angular collection efficiency is described in Section 5. greater attenuation of the illumination and reflected These fibers were led from the collection dome into a light in the medium because the dome is flooded. fiber array block. We then imaged this block onto a The 20-cm compass dome was a reasonable compro- Peltier-cooled CCD camera ͑First Magnitude Corpo- mise between these factors. As the illumination ration͒ using a 16-mm lens system. Each fiber is goes through the dome normal to the surface, and the curvature of the dome is small, there is little change imaged onto a 16-pixel-diameter spot on the camera. in illumination spots between air and water measure- The camera cooling allowed low-noise imaging of the ments because of the refractive index of the measure- fiber block and also a large dynamic range ͑16-bit ment medium ͑dome and air or dome and water͒. dynamic range for digitization͒ for the signal. A me- chanical shutter, provided with the camera, was used B. Viewing Optics to control acquisition time. Along with the viewing The sampling area is in the center of the hemisphere. fibers, a fiber was led from each illumination cavity to An anodized aluminum ring is located at this position the fiber array block. This could be used to verify and helps maintain the sample location and height which illumination fiber was on during the measure- for uneven surfaces. We used 465-m core acrylic ment and provide a convenient normalization of the fibers to view the sample surface. On the instru- data because it was proportional to the light incident ment side of the compass dome a 19-mm-thick dome on the sample. It was also used to check the stabil- of aluminum was machined to hold the illumination ity of the camera, the LED’s, and the operation of the collimation optics and the receiver fibers. Holes camera shutter. 20 November 2000 ͞ Vol. 39, No. 33 ͞ APPLIED OPTICS 6199
- 4. C. System Control average pixel value in this spot is used. Two unil- Control of the system is provided by an embedded 486 luminated portions of the image are used to find an computer. Two two-line displays provide feedback average dark pixel value for the image which is sub- to the diver ͑operator͒ of several instrument param- tracted from the data image during the initial pro- eters ͑instrument temperature, supplied voltage lev- cessing. Therefore the array transferred from the els, maximum and minimum counts in images, file internal computer is the average pixel value for a names, etc͒. A numeric keypad is placed on the given viewing direction with an estimate of the dark front of the case to allow the diver to control impor- value subtracted. Because there are approximately tant parameters during operation of the instrument. 200 pixels averaged to obtain this spot measurement, A 2.5-in. ͑6.3-cm͒ notebook hard drive ͑2 Gbytes͒ random image noise is greatly reduced. When an stores the images in the instrument. A multiple unilluminated sample is reduced, the typical dark- channel analog-to-digital board on the computer al- spot average value is significantly less than one count lows instrument parameters to be measured, and a after this process. digital input-output board controls the line displays After the subsetted array is transferred to the and switches the illumination LED’s on and off. A other computer, the final processing is performed on custom program controls data acquisition, allowing the notebook computer. The first step is to sub- the diver to modify exposure times and illumination tract an average of the six ambient images collected sequences. The whole system is powered from an external from each data set array. This process has two source. We have several choices depending on the effects: It accounts for any light leaks of ambient measurement situation. The instrument is powered illumination into the measurement chamber and it from a 24-V source and requires between 25 and 50 also accounts for residual position-dependent differ- W. Thus we can use a neutrally buoyant instrument ences in the dark values of the spots. The next cable and a dc power supply ͑either a surface battery step is to multiply the data arrays by an array that or wall power if 115 V are available͒. We also have corrects for the relative collection efficiencies of the battery packs that can be attached to the instrument individual viewing channels. A description of the allowing operation independent of the instrument ca- process to obtain these calibration arrays is given ble. This is more convenient for in-water use; how- below. For a given illumination angle and color, ever, it increases the in-air weight of the instrument the viewing channels are divided by the illumina- for laboratory use. tion intensity obtained from the fiber which sam- ples the illumination chamber. Finally the array is multiplied by an absolute calibration number, 3. Data Acquisition correcting for the relative collection and collimation A full measurement sequence consists of three ambi- efficiency of the specific illumination angle and ent images ͑measurements with no LED turned on to color. Thus in the end there are three arrays re- measure the background light leakage͒, 24 data im- quired for data reduction. The first is the associ- ages for the three colors and eight illumination an- ated ambient image, the second is an effective flat gles, and three more ambient images. The total fielding of the viewing channels, and the third is a data set for one sample then consists of 30 images, 750 kbytes͞image or approximately 22 Mbytes͞set. 24-element array with an absolute calibration fac- After a day of measurement, we downloaded the data tor for each illumination angle and color. by transferring the data from the internal computer The later two arrays needed for data reduction are to a notebook computer by a twisted-pair Ethernet obtained during the calibration process. Calibration link. Because the important data are really an ar- of the device is done through comparison with a sam- ray of approximately 110 spots of illumination ͑rath- ple of known BRDF characteristics, in our case a 99% er than a 768 ϫ 512 pixel image͒, data are reflectance Spectralon plaque. We measured the subsampled prior to the transfer process, reducing BRDF of this plaque for normal-incidence light in the data set from 22 Mbytes to 64 kbytes. One full both air and water using a spectral gonio reflectome- image is transferred along with the position file used ter.6 The results of these measurements are shown in the subsampling to allow validation of the data in Fig. 3. The BRDF air measurements were com- reduction process before the raw data are deleted pared with other published measurements of Spec- from the internal hard disk. tralon BRDF, and these measurements were found to agree well.7,8 The in-air and in-water measure- 4. Data Reduction and Calibration ments varied slightly, and the appropriate measure- ment is used during the calibration process. For Data reduction takes place in two stages. The first v Ն 15 deg, the air reflectance values were fit to the stage takes place on the internal computer in the polynomial device, to reduce the size of the data set, before it is transferred to the notebook computer. The image of the fiber array block appears as an array of bright BRDF͑i ϭ 0, v͒ ϭ 0.9686 ϩ 0.001976v spots approximately 16 pixels in diameter. To ab- stract this value for a given viewing direction, the Ϫ 6.470 ϫ 10Ϫ5v2, 6200 APPLIED OPTICS ͞ Vol. 39, No. 33 ͞ 20 November 2000
- 5. 5. System Tests and Characterizations In this section we discuss several tests and charac- terization experiments that we performed with the instrument. We performed many characterization experiments, such as system linearity and spectral calibrations that help define our measurements. In addition we performed tests in which the sample lo- cation varied, which determines the final accuracy of the measurement and the limitations of the tech- nique. These tests are described below. A. Spectral Calibration We measured the spectral output of the LED’s using a spectroradiometer with 1-nm resolution ͑Optronics Fig. 3. Measured Spectralon BRDF values used in the calibra- 740A͒. The red and green LED’s were consistent; tion. Values are given for both air and water ͑submerged͒. The however, the blue LED’s seemed to fall into two Spectralon plaques are the nominally 99% reflectance plaques. groups both in intensity and in spectral emission. Curves are fits to the data described in the text. We selected the group of blue LED’s which gave the greater intensity to be used in the instrument. The central wavelength ͑and standard deviation͒ of the whereas for water the reflectance was fit to the equa- blue, green, and red LED’s were found to be 474.8 tion ͑2.5͒, 569.8 ͑0.8͒, and 657.6 ͑1.8͒ nm, respectively. BRDF͑i ϭ 0, v͒ ϭ 0.9792 ϩ 0.0012770v The FWHM ͑and standard deviation͒ of the output light was found to be 31.7 ͑1.3͒, 28.8 ͑0.3͒, and 24.1 Ϫ 7.430 ϫ 10Ϫ5v2. ͑1.3͒ nm for the blue, green, and red LED’s. Addi- tion of the spectral attenuation of pure water10 sig- These equations each fit the data with a standard nificantly changes only the red LED’s characteristics. deviation of 0.3%. As the water quality changes, the For this LED the central wavelength moves to 654 attenuation of the viewing and illumination beams in nm, and the FWHM narrows to 20 nm. the measurement cavity will change slightly. This is checked by our performing measurements with the calibration plaque in situ. B. Illumination Spot Size For calibration of the device, ten full measurement sets of the Spectralon plaque are made for both air The illumination spots were examined carefully. and water because measurements are desired in both The illumination optics form circular spots on the media. Between each measurement set the Spec- sample surface approximately 1.3 cm in diameter for tralon plaque is rotated 90 deg to eliminate any ori- the normal-incidence spot. As i increases, the spot entation biases of the Spectralon surface. The ten becomes elongated, maintaining the same width but sets of measurements are reduced through the above increasing in length to 3.5 cm at 65 deg. Spot size data reduction process but with the absolute and rel- and location are independent of color because all col- ative calibration arrays set to 1. Because we know ors use the same fiber to transfer light to the colli- the BRDF for normal illumination for this spectralon mation ball lens. The spots from each i also overlap surface we can use the normal illumination measure- on the sample surface and are centered within 2 mm. ments to derive the flat-field array. We assume ͑and tested͒ that the Spectralon surface BRDF is indepen- dent of wavelength over the wavelengths relevant to C. Viewing Fibers our measurement, so we use a combination of the The acceptance efficiency of the viewing fibers was three colors of normal illumination as the flat-field measured as a function of angle in air. These data sample. Once this flat-field array is known the cal- can be fit by a Gaussian function of the form A ibration data are rereduced with this information. exp͓Ϫ͑͞20.64͒2͔. Because of refraction at the The absolute calibration array for the various i window–water interface, the width of this curve is and colors can be determined at this point. We use narrower in water. For water the corresponding fit reciprocity of the reflectance from the Spectralon sur- is A exp͓Ϫ͑͞15.22͒2͔. The size of the dome ͑approx- face, i.e., if i and v are switched, the reflectance imately 20 cm in diameter͒ and the size of the illu- should remain the same.9 We know how the Spec- minated spot ͑1.3–3.5 cm in diameter͒ result in the tralon surface behaves for i ϭ 0°; thus for 0° Յ i Յ illumination subtending a full angle of 7–20 deg for 65° we average v Յ 20° to make the v–i ͑v ϭ 0°͒ the viewing fiber for a view of 0-deg zenith. For this pair with which to compare with i ϭ 0° and 65° Ն geometry the edge of the illuminated spot has a col- v Ն 0°. This can then be used to derive an absolute lection efficiency of 0.97 ͑0.95͒– 0.77 ͑0.64͒ for air ͑wa- calibration number for each color and illumination ter͒. For v Ͼ 0° the viewed area gets bigger, and angle image. the illuminated spot is viewed more efficiently. 20 November 2000 ͞ Vol. 39, No. 33 ͞ APPLIED OPTICS 6201
- 6. with increasing v. For a sample depth of 1.2 cm, this factor ranges from 1.35 to 1.10 as v changes from 5° to 65°, as is shown in Fig. 4. Figures 5–7 illus- trate the residual dependence of the reflectance mea- surement on sample depth, illumination angle, and view angle. Each measurement set at a specific depth was first normalized to the measurement at 0-cm depth ͑the correct position͒. Figure 5 repre- sents the average of all values with v ϭ 5° as a function of i . If this was the only factor affecting the reflectance measurements, Figs. 5–7 should be flat. The additional roll-off of the reflectance with Fig. 4. Correction factor for sample depth as a function of v. sample depth can be accounted for by the change in This factor is predicted by the given sample depth and was applied the spot location and the coincidence of the viewing to the data in Figs. 5–7. spot with the illumination spot. As can be seen in Fig. 5, for i Ͻ 45° and v ϭ 5°, the error is quite small regardless of sample depth. In this case the illumi- D. Variation with Sample Depth nation spot and viewed spot retain their coincidence. Because we chose a geometry in which the illumi- As one increases i to 65°, the spot effectively moves nated area is overviewed, our measurement is depen- off of the center. If the sample surface is 1.2 cm dent on sample height through a 1͞rrs2 dependence. deeper than the correct position, the center of the To investigate this sensitivity to sample location we illumination spot moves over 2.6 cm. This is approx- performed a test in which a Spectralon plaque was imately two diameters of the spot and moves the lowered away from the correct position in steps, and illumination center to a view angle of 15 deg. This a measurement set was collected in each position. drops the collection efficiency of the viewing fiber to The data were corrected for the expected 1͞rrs2 de- 60% and explains the reduction observed in this test. pendence. This factor is dependent on v as it has What this indicates is that the sample must be lo- the greatest effect at v ϭ 0° ͑relative change in rrs cated correctly within 0.5 cm for a measurement er- versus sample depth is greatest here͒ and decreases ror less than 10%. If the distance to the sample Fig. 5. Normalized BRDF and percent error in the BRDF as a Fig. 6. Normalized BRDF and percent error in the BRDF as a function of sample depth and i for v ϭ 5°. function of sample depth and i for v ϭ 35°. 6202 APPLIED OPTICS ͞ Vol. 39, No. 33 ͞ 20 November 2000
- 7. Fig. 8. ͑a͒ System linearity and ͑b͒ percent error from linearity. System is linear over 3 orders of magnitude of response. Multiple data points are the results from each of the fiber viewing spots. Fig. 7. Normalized BRDF and percent error in the BRDF as a function of sample depth and i for v ϭ 65°. F. System Stability As a total system test, a Spectralon calibration re- cannot be estimated, the error is of the order of 20% flectance plaque ͑nominally 99% reflector͒ was mea- from this effect for surface variations of 0.5 cm. sured nine times. Between each measurement the sample was rotated to avoid orientation effects of the E. System Linearity The next portion of the system response to test is the linearity of the system to flux from the sample plane. A thin sheet of diffusing material ͑drafting vellum, 0.6 mm thick͒ was placed at the location of the sample surface, and this diffuser was illuminated from the back by a stable light source. The position of the light source was varied to change the illumination on the back of the diffuser and act as a light source for the BRDF system. The data were extracted for each spot ͑view fiber͒, and these data were normalized to be able to look at the combined linearity for all the view directions. Figure 8͑a͒ shows this combined response versus incident flux. Figure 8͑b͒ is the per- cent error difference between the system response and linearity. The standard deviation of the error in the response from linearity was 5%. Some of the standard deviation is due to how the data were com- bined to form the total data set. As can be seen Fig. 9. Standard deviation of repetitive measurements of a Spec- there is also greater error at the high end of the tralon plaque. The plaque was measured nine times, with a ro- incident flux, where some pixels were starting to sat- tation between each measurement. Data points are the average urate, and at the low end, where dark-noise varia- standard deviation for a whole image, v ϭ 5° and v ϭ 65°. Each tions may be an appreciable portion of the signal. In set of vertical points is for a different illumination angle and LED general the system shows good system response for color. Illumination angle sets are shown on the graph; colors for nearly 3 orders of magnitude of flux. each set vary from red ͑R͒ to green ͑G͒ to blue ͑B͒ from left to right. 20 November 2000 ͞ Vol. 39, No. 33 ͞ APPLIED OPTICS 6203
- 8. Fig. 10. REFF polar plots of a mirror surface. ͑a͒ Polar plot illustrating the location of the view fibers. Each ϩ corresponds to a view direction. The plot is orientated such that the middle ͑0,0͒ is looking normal to the surface. Descending along the y axis corresponds to the specular direction. Ascending along the y axis corresponds to the backscattering directions ͑toward the illumination beam͒. Note the increase of samples in the specular and backward directions. ͑b͒ Normal-incidence, red LED sample shows a large increase in reflectance directly upward. Peak value was 2600%. ͑c͒ i ϭ 25°, red LED shows a large peak in the specular direction ͑1700%͒ and a small peak in the back direction ͑20%͒. ͑d͒ i ϭ 65°, red LED also shows a large peak in the specular direction ͑4300%͒ and a relatively small peak in the back direction ͑100%͒. Spectralon. When the data were reduced, the stan- ilar. If the green LED channels are excluded, the dard deviation was found for each illumination angle standard deviation of the measurements is less than and color. In Fig. 9 the average standard deviations 1%, showing that the instrument can make precise for the total image, v ϭ 5° and v ϭ 65°, are shown. measurements of the BRDF of surfaces. These points are displayed as i and the colors are varied. The first point on the left is the normal il- G. Sensitivity to Spurious Reflections lumination, red LED. The points move from red to As a final test we show Fig. 10, which is the REFF for green to blue as one moves from left to right. Illu- several illumination angles and the red LED. The mination angles are listed on the graph. Several sample measured is a normal front surface mirror. features should be noted in this figure. First, the Figure 10͑a͒ is a polar plot, showing the positions of standard deviations are less than 5% for all cases, the 107 viewing fibers that we have in the instru- and the average for the whole images are less than ment. For orientation, the ͑0,0͒ position is looking 3%. Second, the standard deviation is largest for the straight down at the surface. The illumination green LED. This is the weakest LED, thus the comes in along the vertical direction above the 0° on signal-to-noise ratio is smallest for this wavelength, the y axis. The polar angle moves outward from the increasing the standard deviation between measure- center radially and is proportional to the radial dis- ments. The standard deviation is largest for v ϭ tance from the center. We observe the increasing 65°. This is once again because the signal tends to polar angle, in the specular direction, by moving ver- be less at large v, causing a decrease in the signal- tically downward ͑negative angles along the y axis͒. to-noise ratio. There is a slight increase in the stan- Contours for this sample are at every 0.02 for dard deviation with increasing illumination angle, REFF Ͻ 0.1, 0.1 for 0.1 Ͻ REFF Ͻ 1, 1 for 1 Ͻ but this trend is not clear. Finally the standard de- REFF Ͻ 20, and 5 for 5 Ͻ REFF. Figure 10͑b͒ il- viations for the red and blue LED channels are sim- lustrates the REFF͑i ϭ 0°, v, ͒ for the red LED. 6204 APPLIED OPTICS ͞ Vol. 39, No. 33 ͞ 20 November 2000
- 9. Fig. 11. REFF polar plots of dry and wet sand surfaces. ͑a͒ REFF ͑i ϭ 0°, v, ͒ at 658 nm for a dry sand surface. ͑b͒ REFF͑i ϭ 65°, v, ͒ at 658 nm for a dry sand surface. ͑c͒ REFF͑i ϭ 0°, v, ͒ at 658 nm for a wet sand surface. ͑d͒ REFF͑i ϭ 65°, v, ͒ at 658 nm for a wet sand surface. As one would expect with a mirror, there is an in- 6. Example Data crease in the reflectance in the specular direction, As an example of the measurements that we can ob- and the reflectance is fairly symmetric around this tain with the system, we show two examples. The direction. The peak value of the reflectance in this first example is a sample of dry construction-grade case was 26 ͑note that 1 is relative to a diffuse sur- sand. This sand appeared white when dry. Figure face͒. Figure 10͑c͒ is the REFF͑i ϭ 25°, v, ͒, also 11͑a͒ shows the REFF͑i ϭ 0°, v, ͒ for 658 nm. The for the red LED. In this example the data show two orientation of the graph is the same as in Fig. 10. The peaks. The first is the specular peak ͑peak value contours in Fig. 11͑a͒ are at 0.02 for REFF Ͻ 1, 0.1 for observed was 17͒. Outside this specular peak the 1 Ͻ REFF Ͻ 2, and 0.2 for 2 Ͻ REFF. As can be seen, reflectance is low; however, there is a peak in the for normal illumination, the sand sample is diffuse and backward direction of 0.20. This peak is approxi- Lambertian to within Ϯ0.02. The lack of contours mately 1% of the specular peak and is due to back- reflection from the window of the specular peak back over most of the image indicates the uniformity of the down to the mirror and up to this view direction. sample. The nominal reflectance of the sample is This is the worst type of sample for this effect, and 0.30 Ϯ 0.01. This REFF can be contrasted with Fig. still the error that is due to this retroreflection is only 11͑b͒, which is the REFF͑i ϭ 65°, v, ͒ for 658 nm. of the order of 1%. Figure 10͑d͒ is the REFF͑i ϭ This sample is also dry sand, but with a large illumi- 65°, v, ͒, also for the red LED. Once again the nation angle. This sample illustrates two effects of- specular peak is very high ͑43͒, and the retroreflec- ten seen in natural samples. First there is a fairly tion is of the order of 2–3% of the specular peak. large specular peak. Although the reflectance, nor- This data set illustrates that the instrument error mal to the surface, is still approximately 0.32, the spec- that is due to internal reflection is very low, but must ular peak is 0.90. This sample also has an be considered for intensely specular surfaces. Note enhancement in the backscattering direction. In this that in water this effect would decrease because of direction the peak is approximately 0.70. This is the decreased reflectance at the water–window interface. effect of the hot-spot commonly seen in BRDF’s of var- Also a diffuse reflecting surface would spread the ious surfaces.3 At these larger angles of illumination retroreflection down to the noise level. the samples are clearly not Lambertian. 20 November 2000 ͞ Vol. 39, No. 33 ͞ APPLIED OPTICS 6205
- 10. The next sample shown is actually the same dry This instrument will be used to make field measure- sample measured above, but water was added to the ments of the BRDF in various environments to help sand to make the surface wet, filling the pores with characterize benthic surfaces and to improve model- water. This is not a submerged sample, but we ing and remote sensing algorithms. changed the surface by adding water to the sand. Figure 11͑c͒ is the REFF͑i ϭ 0°, v, ͒ for 658 nm. The research presented in this paper has been sup- First it is obvious that the overall reflectance of the ported by the Ocean Optics Program of the U.S. Office surface has changed significantly. In this case the of Naval Research ͑N001499-10008͒. We thank nominal reflectance of the surface is 0.16 Ϯ 0.01. Karl Moore who helped in the early development of Once again the surface is Lambertian at normal in- the instrument. cidence. This is again contrasted with Fig. 11͑d͒ that is the REFF͑i ϭ 65°, v, ͒ for 658 nm for the References same sample. Once again, when the surface is wet 1. B. Hapke, Theory of Reflectance and Emittance Spectroscopy, the normal reflectance is lower than in Fig. 11͑b͒. Vol. 3 of Topics in Remote Sensing ͑Cambridge U. Press, New As in Fig. 11͑b͒, there is a large specular peak. In York, 1993͒. fact, in this case the reflectance of the specular peak 2. R. A. Leathers and N. J. McCormick, “Algorithms for ocean- has increased to 1.60, which is much larger than in bottom albedo determination from in-water natural-light mea- the dry sand. This is probably caused by having surements,” Appl. Opt. 38, 3199 –3205 ͑1999͒. some of the surface covered with water and is a re- 3. J. M. Chen and J. Cihlar, “A hotspot function in a simple bidirectional reflectance model for satellite applications,” J. flection from the air–water interface. Measurement Geophys. Res. 102, 25907–25913 ͑1997͒. of submerged sand samples have not shown this large 4. D. R. White, P. Saunders, S. J. Bonsey, J. van de Ven, and H. specular peak. The backward-scattering peak has Edgar, “Reflectometer for measuring the bidirectional reflec- diminished significantly from that seen in Fig. 11͑b͒. tance of rough surfaces,” Appl. Opt. 37, 3450 –3454 ͑1998͒. Here the peak is only 0.24. Although this is greater 5. J. R. Zaworski, J. R. Welty, and M. K. Drost, “Measurement than the reflectance at v ϭ 0, it is still not as large an and use of bi-directional reflectance,” Int. J. Heat Mass Trans- increase as seen in the dry sand. fer 39, 1149 –1156 ͑1996͒. 6. J. W. Giles and K. J. Voss, “Submerged reflectance measure- 7. Conclusion ments as a function of visible wavelength,” in Underwater We have described a new instrument to make rela- Imaging, Photography, and Visibility, R. W. Spinrad, ed., Proc. tively rapid in situ measurements of the BRDF of SPIE 1537, 140 –146 ͑1991͒. surfaces. This instrument will allow the BRDF for 7. B. T. McGuckin, D. A. Haner, and R. T. Menzies, “Multiangle many surfaces to be measured and characterized for imaging spectroradiometer: optical characterization of the use in optical modeling of the light field and remote calibration panels,” Appl. Opt. 36, 7016 –7022 ͑1997͒. sensing. Our major concern when using the instru- 8. K. D. Moore, K. J. Voss, and H. R. Gordon, “Spectral reflectance of whitecaps: instrumentation, calibration, and performance ment is to make sure that the sample height is correct in coastal waters,” J. Atmos. Oceanic Technol. 15, 496 –509 because the accuracy is dependent on this factor. ͑1998͒. The precision of the measurement has been shown to 9. W. C. Snyder, “Reciprocity of the bidirectional reflectance dis- be 1–5%. The accuracy of the measurement is de- tribution function ͑BRDF͒ in measurements and models of pendent on how well the surface height has been structured surfaces,” IEEE Trans. Geosci. Remote Sens. 36, determined. If the surface is correctly placed within 685– 691 ͑1998͒. 0.5 cm, the accuracy is approximately 10%, depend- 10. R. M. Pope and E. S. Fry, “Absorption spectrum ͑380 –700 nm͒ ing on v and i ͑note that this is 10% of the measured of pure water. II. Integrating cavity measurements,” Appl. reflectance value, not 10% in reflectance units͒. Opt. 36, 8710 – 8723 ͑1997͒. 6206 APPLIED OPTICS ͞ Vol. 39, No. 33 ͞ 20 November 2000