Change Detection in Colour :
Presentation and Interpretation of Multi-Dimensional Image Data Sets
Anthony J.
Lewis1,3, Qiang Tao2 and DeWitt H.
Braud1 Abstract1Department of geography and Anthropology Louisiana State University, Baton Rouge, LA USA 70803 Tel:504-388-6199 Fax:504-388-2912 E-mail: ajewis@cypress.cadgis.Isu.edu 2Coastal Studies Institute Louisiana State University, Baton Rouge, LA USA 70803 Tel:504-388-4429 Fax:504-388 E-mail: qtao@cypress.cadgis.Isu.edu 3Senior Fellow, Department of Geography, The National University of Singapore, SINGPORE 119260 (From 12/96 to 12/97), E-mail: geoal@nus.sg The perception and interpretation of colour enhance our understanding of the environment in which we live. Colour is used to present multi-spectral reflectance patterns in a colour television or standard colour schemes when energy outside of the visual range is recorded. In either case, the images generated utilize three layers of colours to present multi-spectral information. By incorporating the same colour theory and technology multi-dimensional false colour composite images that show change (spectral, temporal, polarization, look angel etc.) in a spatial context can be generated. To interpret these multi-dimensional false colour images correctly the interpreter must be cognizant of 1) the sensor parameters 2) the assigned colours, 3) colour theory and 4) the signal-target interaction. A standard procedure for assigning colours is recommended. Introduction Colour is a primary visual key to the perception and interpretation of our environment. Colour and its use are undoubtedly important contributors to a better understanding of the environment in which we live. Colour is defined by both physical laws and physiological perceptions. Physical laws describe colour in terms of wavelength and frequency; physiologically colour is described in the way that each of us" perceives" a certain colour. Description by physical laws is precise. In even through it may be subtle difference. To help reduce some of the subjectivity in using colour to identify and classify features, standardized colour chips, e.g. Munsell Colour Chart, are used as a reference source. Soil scientists are very familiar with Munsell Colour Charts and commonly use them to describe soil colour characteristics. Primary Colours Colour can be divided into 1) Additive primaries and 2) Subraractive primaries. The three additive primaries are Blue, Green and Red. The combination of these three additive primary colours results in white light. Subtractive primary colours are the result of white light minus one of the additive primaries. Although subtractive primaries are more logically thought of as white light minus an additive primary, subtractive primaries are also the combination of two additive primaries. For example, white light minus blue light results in the subtractive primary yellow. Yellow can also be described as red plus green. Both types of primaries have special characteristics and are used different applications for colour. Additive primary colours are used to generate colours in colour TV's and colour monitors for computers, etc. Colour TV's utilize three electron guns to excite the blue, green and red phosphors on the monitor. In he photographic process, for both the production of slides and prints, subtractive colours must be used. With both slides and prints, the light that reaches the eye and therefore carries the colour information is filtered by the colour print. Colour and colour theory are complex topics and beyond the scope of this paper. For more detailed discussions of colour and colour theory the reader is referred to Kueppers (1982) and Nemcsics (1993). Colour and Colour Infrared Film All colour films, slides and prints, regards of the manufactures, are tri-emulsion films (three layers of colour dyes). Standard or "natural" colour film has three layers, each layer sensitive to one of the additive primary colours. The dye assigned to each of these layers is the complementary subtractive primary colour. For example, a yellow dye is assigned to the blue sensitive layer. Although the colours on "natural" colour film may be different on different films, they are close to that of the original scene. Blue light reflect from the feature and recorded on standard colour film appears blue in colour. Reflected green light appears green, and reflected red light appears red. When the assignment of colour days is not conventional, colours exhibited on the film are "unnatural." Such film is generally referred to as "false colour film." Colour infrared film, perhaps the most common type of "false colour film", has green, red and near-infrared sensitive layers as the three standard layers. Since these layers also exhibit sensitivity to blue light, an external yellow filter is usually attached to the lens of the camera to eliminate incoming blue light. When exposed and developed the green sensitive layers will appear blue, the red will appear green and the near infrared will appear red. If a object only reflects blue light, it will be black on colour infrared film. For both conventional natural colour film and false colour infrared film, colour is used to present spectral differences in a scene taken at single point in time. The colours on the photograph provide the viewer (photo interpreter) with information on the way the object (s) absorb and reflect energy within the spectral bands being recorded. Since most photographs are captured using solar energy infrared energy from the sun. Therefore all colour film is a record of multi-spectral reflectance values collected at an instantaneous point in time. For a more detailed description of colour and colour infrared film the reader is referred to almost any introductory test on remote sensing, e.t. Avery and Berlin (1992) and Lillesand and Kiefer (1994). Multi-Dimensional False Colour Images The discussion so far has focused on a brief explanation of colour and the different types of colour films. It is important to have a basic understanding of these concepts before attempting to visualize the presentation of other multi-dimensional data sets using colour. Using colour to view multi-spectral data is based on the same principles used to present other types of data sets (temporal, polarization, look direction etc.). Three different but spatial registerable data sets are dimensional models are presented below : Multi-spectral model - S1T1; S2T1; S3T1 Multi-temporal model - S1T1; S1T2; S1T3 Multi-polarization model - S1P1; S1P2; S1P3 Multi-look angle model - S1L1; S1L2; S1L3 Multi-look direction model - S1D1; S1D2; S1D3 where S represents a spectral band; T represents a given point in time; P represents polarization; L represents look angle and D represent look direction. In all of these models there are only two parameters presented: one remains constant and the other varies. The reader should be aware that for a better understanding of the signal-target interaction, only one parameter should be varied in each model although in reality each model has more than two parameters. For example, for interpretive purposes time (T) should only be varied in the multi-temporal model and is assumed to be constant in all of the other models. The "multi-spectral model" is commonly utilized in conventional" normal or true" colour images and false colour image. The "multi-temporal model" is another common model. All multi-temporal images are false colour images. As indicated above, a false colour multi-temporal image is the composite of a a single spectral band collected at three different time periods (S1T1; S1T2; S1T3). Each of the three time periods of information are assigned colours just as in the production of a multi-spectral colour photograph. In this case, however, the colours represent change in brightness values or tone (spectral reflectance or thermal emittance) over time. The change recorded may be due to a variety of variable that can affect reflectance or emittance (sun angle, atmospheric conditions, land cover changes, etc.). The challenge for the interpreter is to make sense of the colours and to isolate the parameters affecting the change. The specific colours on the multi-temporal false colour image are also ependent on the colours assigned to each of the three time period. For example, the scenario of additive colours used in a colour monitor presentation will determine - along with actual change in reflectance - the colours on the final product of a multi temporal false colour image. if the colours blue, green and red are used for the temporal sequence temporal false colour image.If the colours blue, green and red are used for the temporal sequence T1, T2 and T3, respectively, a change from high (white) to low (black) brightness values between T1, and T2 and remaining low (black) in T3 would result in the colour blue; whereas, if the colour for sequence red, green and blue were used for the same time sequence, T1, T2 and T3 the colour for the same brightness value change would be red. By standardizing the colours assigned to the time sequence to T1, (blue) T2 (green) and T3 (red) the meaning of the basic colours on a multi-temporal false colour image can be summarized (See Table 1 and 2). Only chromatic colours (blue, green, red, cyan, magenta and yellow) indicate change (Table 1); achromatic colours (white, black and grey) indicate no change in reflectance over time (Table 2). Table 1. Reflectance change over time indicated by chromatic colours on multi-temporal false colour image.
With knowledge of the spectral bands selected for the multi-temporal false colour composite and an understanding of the spectral reflectance and /or thermal emittance characteristics of the target, the image analyst may be able to decipher the meaning of the colours and determine the reason (s) for the tonal change over time. For example, if three dates of thermal imagery were composited using the colour scenario presented in Table 1, a feature coloured blue would be interpreted as being hot in T1 and cold in T2 and T3. If near-infrared imagery (for example, Landsat TM, Band 4) were used to generate a three data multi-temporal false colour composite of a coastal or wetland area, the colour blue could indicate a change from land to water (land loss) during T3 and T2 and remaining as water in T3. However, if the area imaged was dominated by agricultural by agricultural patterns, the same tonal change and colour sequencing scenario could indicate a healthy crop in the field during T1, and the field in fallow during T2 and T3. Table 2. Achromatic colours on multi-temporal false colour image indicating no reflectance change over Time
Colour Radar Colour radar has been around for more than twenty years; however, until recently the image were "colourised radar and not " true" colour radar. Early "colourised" radar images were single band radar image that were level sliced, and colours were assigned to selected ranges of gray scale values (digital numbers (DNs) or brightness values (BVs(). Colour composite radar images, on the other hand, use multi-dimensional radar data sets and are generated by assigning a primary colour (blur, green or red) to the three single dimensional data sets being combined. If registerable, any two of three different sets of multi-dimensional data sets can be combined in his way to generate colour composite radar images or, perhaps more accurately, a false colour multi-polarized, multiple incident angle, multiple look direction, and multi-squint angle radar data. Examples of false colour multi-dimensional radar images are found in a variety of NASA publications (Ford et al., 1989; NASA 1986; 1989) and ESA publication, such as esa bulltetin. False colour multi-dimensional radar images are also available through the NASA/JPL home page (southport.jpl.nasa.gov) and other home pages on the Internet. Although radar or active microwave data can be used to generate multi-dimensional false colour images, the interpretation of the colours is not as "conventional" as with visible and infrared images because of the differences in the signal to target interaction. Tonal (brightness) changes on radar imagery are related to system and target related parameters defined in the radar equation as well as to the interplay of these system and target parameters (Table 3). Complex interactions and relationships take place between and among the parameters in Table 3 when the signal penetrates the surface and volume scattering occurs. In addition to the parameters in Table 3, complex volume scattering, subsurface roughness and the index of refraction must be taken into account, often in a very convoluted manner. For example, the amount of radar penetration is a function of wavelength, complex dielectric, incident angle and polarization coupled with surface and subsurface roughness characteristics and the medium's index of refraction. Table 3. Foundamental system and target parameters that influence rader brightness (power return)
In order to simplify the interpretation of a radar image, the target parameters need to be isolated from system and imaging parameters (frequency, polarization, took direction and look or incident angle) and, if possible, other target parameters, that will affect radar backscatter and mask colour radar composite can be straight forward or nearly impossible to discern depending on the parameters used in collecting the data. In order to improve the interpreter's understanding of the "reasons" for the colours, the system and imaging parameters should be kept constant when composting multi-temporal false colour imges. If the interest is in other facets of multidimensional radar data (frequency, polarization look direction, incident angle etc.), then time and all but the interested parameter should remain constant. Multi-band radar images, and in fact all multi-dimensional false colour radar images, are to radar remote sensing what multi-spectral photographs or images are to black and white (single band) photo interpretation. That is, there is a significant increase in the amount of spectral information about the target when more then a single band or l is used. The analyses of false colour multi-dimensional radar images are providing a better understanding of surface roughness characteristics, as well as the influence of wavelength, polarization and incident angle. With the successful acquisition of multi-dimensional radar data by NASA AIRSAR and SIR-C/S-SAR and the increasing availability of the data to geoscientists, the generation of multi-dimensional colour radar image has commenced and analysis is underway. In an attempt to standaridize the presentation of these false colour radar images, it is recommended that colours assigned to the radar bands follow the same sequence as colour photography. Blue is assigned to the shortest wavelength; green to the middle wavelength; and red to the longest wavelength. Suggestions for standardization of multi-dimensional false colour images are presented in Table 4. Conclusions In addition to understanding colour theory, it is imperative that the image analyst know 1) the type of data set being combined; 2) the colour assignment scheme; 3) the number of data sets of layers of information being combined, and 4) the physical processes that influence spectral reflectance and emittance as well as radar backscatter. A lack of knowledge or understanding of any of the above will usually result in misinterpretation of the data set. As in al cases of image analysis, the more interpreters know about the above parameters in addition to their knowledge of the study area and the applicable scientific discipline (geology, geography etc.) the ore accuracy the interpretation will be. In addition, interpreters should not base the final analysis entirely on tonal change but should not base the a final analysis entirely on tonal change but should utilize as many of the other interpretative keys (texture, pattern, size etc.) as possible, allowing for a convergence of evidence. Table 4. Suggested standarized colour scheme for multi-dimensional false colour (fc) radar images and other remotely sensed images.
The use of colour to present multi-dimensional data enables the interpreter to visualize multi-dimensional data (spectral, temporal, polarization etc.) and changes in a spatial context on a single image. Incorporating a standardized colour scheme, such as in Table 4, will 1) reduce another concern for the interpreter 2) help provide a better understanding of the signal/target interaction and 3) allow for more comparative analyses between colours of different scenes or multi-dimensional false colour composites from a variety of sensors and situations. Bibliography
|