The Use of lntensity'Hue-SaturationTransformation for Producing Golor Shaded'

Relief lmagesKathleen Edwards and Phlllp A. Davls

AbstractWe present a novel technique for combining c-olot infotya-tion'from one raster dataset with intensity information fromonoih", by using an intensity-hue-saturation (tus) transfor-mation. ihis teihnique, which we refer to as intensity adjust-ment, is rclated to IHS trunsformation methods (e.9.' intensitysubstitution) that have recently been used to combine re-motely sensed multispectral images with ach,tomatic images -of higher spatial rcsolution. Our intensity-adiustment methodii nol oppiopriate for merging multispectral and achromaticimages'whoie intensity information is mostly redundant, butourlechnique is more effective than the intensity-substitutionmethod in merging databases whose intensities arc nonte-dundant, such as a color thematic map and a shaded-reliefbase image.

IntrcductionThe basic concept for transformation of a digital elevationimage into a shaded-relief image was first explained by Bat-son-et al. (f Szs;. Since that time most image-processing soft-ware systems have incorporated this basic technique arrd -expanded on it so that three-dimensional renditio-ns of a dig-itai elevation image can be produced in which different vi-sual effects are achieved by varying the values of suchparameters as surface/sun orientation, sun-elevation angle, _ind vertical exaggeration. We have taken another step in thisprocess by developing an algorithm to combine a color-ioded digjital imafe with a digital shaded-relief image. Thistechnique preserves both the distinct colors in the thematicimage ind the three-dimensional brigltness information ofthe ihaded-relief base image. The method was designed toconvey instantly the spatial relation between topography and

".r uniilla"y thernatic image (a map) of a particular attribute

of a geographic region. Examples of such images that would

"omfl"-"ttt a three-dimensional topographic image are those

showing magnetics, gravity, elemental or mineral c-omposi-tion, geology, population density, landslide or earthquake-risk, surface-water quality, and a Landsat ratio. The uses forthis algorithm are limited only by one's imagination.

BackgfoundTwo of the more common ways to represent color are (1)

designation of the digital numbers (oNs) of the three primarycolois (red, green, and blue - ncn), which are used to pro-duce color images on color monitors and photowrite scan-ning systems, ana (z) designation of intensity, hue, and

saturation (ms), which are used to define color mathematically-in cylindrical or spherical coordinate systems. An advantage of

"o.tJid"ritrg color in terms of IHS over that of RGB is that the

manner an"d effects of combining two data sets can be evaluatedand adiusted mathematically to arrive more easily at a desiredcolor product.

One application of the IHS system has-been in merginghieh-resolition achromatic imagb data with lower resolutioncolo, image data. The modified color image shows the spatialdetail of tie achromatic data while maintaining much of thespectral content of the original color data (Cliche et o/', 1985;lielch and Ehlers, 1987; Carper ef o1., 1990). This digitalmerqer is accomplished by calculating a pixel's IHS valuesfroni its Rcs values (Plate 1a), substituting for that intensityvalue the intensity of the corresponding pixel in the achro-matic image (Plate rb), and then recalculating the pixel's newRGB valuei hom its modified IHS values (Plate rc). To mergedigitally two images that have different pixel.resolutions, it isfirit neiessary toiegister the data geometrically; such registra-tion procedures have been disc-ussed by Chavez (1986)'

'ih" i*"g"--erging methods reviewed-and evaluated byCaroer et o1. irsso)-work best when the achromatic data areredindant with the color data, i.e', when the wavelength re-gion represented by the achromatic data closely- approximates[he wavelength range represented by t]re color data' In this sit-uation the iritensities oflhe achromatic image and the RGB im-age are very similar when expressed as normalized intensitiesbised on their maximum oN iange. If one of the RGB imagefiles is not redundant with the aihromatic image file' as whensPor panchromatic data (0'51 to 0.70 pm) are merged-withspor isr (0.s0 to 0.59 pm), xsz (0.61 to 0.68 pm), and xsr(0.79 to O.-ag pm) data, then the substitution intensity needs toLe calculated as a weighted average of the achromatic dataand the nonredundani(xss) band-data (Carper et o1', 1990)'This problem is magnified when all three color-image files arenot redundant with the achromatic image base (e'g', merging acolor-coded thematic map with a shaded-relief base image;Plates 2a, 2b, and 2c). In this PaPer we describe a method todigitally merge color-coded th-ematic images- with a shaded-re-liei basL imaEe; this method retains, as much as- possible, boththe colors of'the thematic image and the three-dimensional as-pect of the shaded-relief base image.

Photogrammetric Engineering & Remote Sensing,Vol.60, No. 11, November 1994' pp. 7369-7374.

0099-1 1 1 2/94/601 1-1 369$3'00/0@ 1994 American Society for Photogrammetry

and Remote SensingU.S. Geological Survey, 2255 North Gemini Drive, Flagstaff,

AZ 8600L

PE&RS

Plate 1. An agricultural and suburban area near phoenix, Ari-zona. (a) Landsat TM falsecolor composite image (3Gm resolution) of TM bands 1 (displayed in blue), 3 (displayed in red),and 4 (displayed in green). (b) SPOT panchromatic image (1Gm resolution; copyright SPOT data CNES). (c) Landsat TM andSPOT panchromatic images lH$merged by intensity substitu-tion (cliche et al., 1985).

IHS TranshrmationThe concepl of intensity, hue, and saturation was reviewed byCarper ef d1. (1990), and the geometric transformation of apixel from RGB coordinates to IHS coordinates was describedby Siegal and Gillespie (1980, pp. 2O3-ZOS). There are differ-ent mathematical representations of this transformation, de-pending on the primary color used as the reference point for

1370

hue and on the manner in which intensity is calculated. Wehave chosen blue as our reference point for the IHS coordinatesystem- (Figure 1), and we represent a pixel's intensity as thesum of its RGB otts. The following equations relate a pixet,sRGB DNs to IHS values in cylindrical coordinates along the ach-romatic axis:

PE&RS

Plate 2. North and Central America. (a) Colorcoded topegrapnic image. Many of the elevation intervals have the samehue but different intensities. (b) Shaded-relief image. (c) Color-coded and shaded-relief topographic images lHgmerged by in-tensity substitution (Cliche et al., 1985).

I = ( D N R + o N . + D N " ) (1) Conversely, the following equations relate a pixel's IHSvalues to RGB DNs:

_ r scos(n) ss in(H)

V6 \/,

^ r scos(H) ss in(H)

V6 -

\/2

(2)H:tan-1 t#.#'*]

(3)

Blue

Red

Figure 7. RGB and IHS coordinate system (adapted fromSiegal and Gillespie (1980)),

0.5, r"", will be less than i*. Conversely, if r** is greaterthan 0.5, i"", will be greater than ir. If r'*" equals b.5, i.",will equal r,".

Step 5. Calculate a new color-intensity value (Iu"") by multiply-ing the new normalized intensity (r"*) by ZOS.

Step 6. During this procedure we preserye the original ratio ofs.d8ls-*(t"d'), where Son" is the original saturation valueand s-*(t) is the maximum possible saturation value fora given intensity, by determining s-*(tu"J and multi-plying it by S.dBls.-{I.d8) to determine t}re new satura-tion value.

Step 7. Calculate the pixel's new RGB values by using its newintensity and saturation values and its original huevalue from the color image.

Step 8. This procedure can result in RGB oxs that are less than0 or greater than 2b5. We therefore check these twoconditions for each of the RGn DNs consecutively, andwe adjust the RGB oNs so that their values are withinthe range of 0 and 2bS (see Appendix).

ResultsOur intensity-adjustment method was applied to two differ-ent types of data. The first application involved redundantachromatic (seor panchromatic) and color (seor multispec-tral) data (Plate 3a); the second application involved nonre-dundant achromatic shaded-relief and color-coded elevationdata (Plate 3b). In the first application the intensity-substitu-tion method (Plate fc) produced a color image superior tothat of the intensity-adjustment method (Plate 3a). Thus, theintensity-substitution method should be used with redundantdata. However, in the second application, which involvednonredundant data, the intensity-adjustment method (Plate3b) produced a color image superior to that of the intensity-substitution method (Plate 2c). Thus, our intensity-adjust--ment method should be used with nonredundant-data.However, redundancy of the achromatic and chromatic datais not the only issue.

- In deciding between intensity substitution and intensityadjustment for image merging, one must also consider the

-

hequency distribution (histogram) of the intensity modula-tion in t}le achromatic image. The visual cortex in the hu-man eye consists of two cell types (magnocellular genticulateand parvocellular genticulate); the magnocellular cells aremost sensitive to intensity, whereas the pawocellular cellsate most sensitive to color information (Livingstone and Hu-bel, rsea). These different cell responses caus; these twotypes of information to follow different signal-processingpaths: the color information is processed at lower resolulionthan the intensity information. (fhis is why a black-and-white photograph appears to have more clarity than a colorphotograph.) Thus, if the intensities of the achromatic imageare concentrated at either the low or high end of the inten-sity range, or at both ends, as in the case of suburban areas,the intensity-substitution method is better, because it trans-fers more of the intensity information of the high-resolutionachromatic image to the merged color image, to which theeye is more sensitive anyway. On the other hand, if the in-tensities of the achromatic image are concentrated at themiddle of the intensity range, as in the case of a continentalshaded-relief map, the intensity-adjustment method is prefer-able, because it will better retain the intensities of the iolorimage.

AcknowledgmentsThe authors are grateful to Alex Acosta and Randolph Kirkfor their thoughtful reviews. This research was supported as

(6)- I s V O c o s ( H )l J : - - l -

3 3

The intensities of a shaded-relief image and a color-coded thematic image are generally not redundant, so thatapplication of an intensity substitution method (e.g., Clicheet d1., 1985) to these data will produce intensities noticeablydifferent from those in the original color image (Plates 2a,2b, and 2c). To maintain most of the intensities of the origi-nal color thematic image during the process of combiningthe shaded-relief base with the color thematic data. the in-tensities of the color data should be adiusted toward, ratherthan substituted for, the intensities of the shaded-relief im-age.

We accomplish this intensity adjustment, for each pixel,by the following procedure:

Step 1. Convert the RcB oNs to IHS values.Step 2. Convert the color intensity value (r.n") to normalized in-

tensity (4") by dividing t.d" by 765, which is the maxi-mum intensity value that can be obtained by summingthe RGB DNs. We likewise convert the shaded-relief ottto a normalized intensity (ri"*) by dividing its ou by255.

Step 3. Establish a piecewise linear relation between the nor-malized intensities of the input color (r,.) and the de-sired output color (r",,). This transformation is tied tothe pixel's 4"". value: it will generate an r"", value equalto r0"". when i," is 0.5. The i"-r"", transformation is alsodefined so r," of o and 1 remain unchanged. This inten-sity transformation or adjustment can be expressedmathematically as follows:

,"", : {iT"._l?"':i,,,,^ * zib^",, _ 1, ' jBl 0)

Step 4. Calculate a new normalized color intensity (1"",) basedon t," and r0""" and using Equation 7. If ib.". is lower than

L372 PE&RS

Plate 3. Examples of |Hs-merged data using our intensity adjustment method on (a) Landsat TM (Plate 1a) and SPOT panchromatic (Plate 1b) data, and on (b) colorcoded (Plate 2a) and shaded-relief (Plate 2b) topographic images.

part of various contracts hom NASA's Solar System Explora-tion Division.

RefelencesBatson, R. M., K. Edwards, and E. M. Eliason, 1975. Computer-Gen-

erated Shaded-Relief Images, I. of Research Il.S. Geoi. Survey,3(4) :401-408.

Carper, W. ]., T. M. Lillesand, and R. W. Kiefer, 1990. The Use ofIntensity-Hue-Saturation Transformations for Merging SPOTPanchromatic and Multispectral Image DaIa, PhotogrammetricEngineering & Remote Sensing, 56(4):459-467.

Chavez, P. S., 1986. Digital Merging of Landsat TM and DigitizedNHAP Data for 1:24,000-Scale Image Mapping, PhotogrammetricE n gi nee ri n g & Re mote S e n s i ng, 52(1. O) :76 37 -76a6.

Cl iche, G., F, Bonn, and P. Tei l le t ,1985. Integrat ion of the SPOTPanchromatic Channel into its Multispectral Mode for ImageSharpness Enhancement, Photogrammetric Engineering & Re-mote Sensing, 51(3):31 1-316.

Livingstone, M., and D. Hubel, 1988. Segregation of form, color,movement, and depth: Anatomy, Physiology, and Perception,Science, 24O:74O-749.

Siegal, B. S., and A. R. Gillespie, -198O. Remote Sensing in Geology,Wiley, New York, 702 p.

Welch, R., and M. Ehlers, 1987. Merging Multispectral SPOT HRVand Landsat TM Data, Photogrammetric Engineering & RemoteSensing, 53(3):301-303.

(Received 28 December 1992; accepted 8 April 1993; revised 20Apr i l 1993)

AppendixOur intensity-adlustment technique may result in RGB DNs (instep 7) that are less than 0 or greater than 255. The following

PE&RS

mathematical sequence is used to ensure that the final ncgDNs are within the range of 0 and 255. The symbol I that isused in these equations represents 16""r.

i fR < 0 , se tR = 0

B : I - Gi f c < 0 , s e t c : o

R : l - Bi f B < o , s e t B : 0

G : l - Ri f R > 2 5 5 , s e t R : 2 5 5

B :i f c > 2 5 5 , s e t G :

R :i f B > 2 5 5 , s e t B :

I - c - 2 5 5255

l - B - 2 5 5255

G : I - R - 2 5 5

the Earth and planetstechniques that she has pioneered include the geometriccontrol and reprojection of spacecraft images, mapping ofirregular objects (small satellites and asteroids), and the au-tomatic generation of shaded relief maps ftom digital topo-graphic data. Ms. Edwards has participated in numerousplanetary missions, including Viking, Voyager, and Galileo.In 1989 she coordinated the production of a global digitalimage mosaic (olv) of Mars from over 4500 Viking Orbiterimages, and helped to create a high-resolution color oIv of

Kathleen EdwardsKathleen Edwards received her B.S. in Math atNorthern Arizona University in 1969. She hasworked at the U.S. Geological Survey since1964. Since 1975 she has worked as a1964. Slnce 1975 she has workecl as a program-mer, creating software for digital cartography ofI planets using spacecraft data. The specialized

Neptune's satellite Triton in real time during the Voyager 2encounter.

Philip A. DavisPhilip A. Davis received his B.S. in Geology atBowling Green State University (1972), M.S. inGeology at Miami (Ohio) University (197a), andPhD in Geology at University of Kentucky

r1,,,;;,''r (7572). He began remote-sensing research as apost-doctoral research chemist at U.C. San Diego in 7977 bymapping elemental distributions on t}re Moon using Apolloorbital gamma-ray data. Since 7977 he has worked at theU.S. Geological Survey developing and applying image proc-essing techniques in the extraction of topography using mon-oscopic images of planetary surfaces and in geologicmapping and mineral exploration using Landsat MSS andTM and radar data of various semi-arid and arid regions ofthe world.

Digital Photogrammetry in an IntegratedProduction Environment/4SPR^S Worl<shops in the St. Louis Regton2nd and 3rd December 1994Holiday Inn Riverfront, St. Louis

. Implementing and Managing of GISRebecca Somers of Somers-St. Claire, Fairfax, VA

o Digital Imagery for GIS and Mapping ApplicationsIim Linders of the University of Guelph, Ontario

o Issues and Concepts in the Design and Implementation ofa Digital Stereoplotter

Craig Molander, Wsion International and Brian Cabral, SGI

o Panel Discussion: Integration of Digital Photogrammetryinto a Production Environment

To receive a completedescription of the session

and details on some terrific rates andaccommodations, leave your address and number

at ASPRS HQ, 301-493-0290 x20 or fax 301-493-0208

L374 PE&RS