Property Maps Manual

At this time, there are four general categories of map types:

The majority of maps use an image file as input. DIRSIG supports a number of common image formats. There is no need to specify the specific format of the file, it will be interpreted from the format extension, or alternatively from from signature of the file itself. The supported formats (with the standard extension) are listed below:

Common to both projection methods is a set of options describing how the source image file is to be mapped from image to projection coordinates and how to handle extension of the image when the projection maps to pixel coordinates outside of the image range.

The tile (or "drape") projection simply maps two dimensional pixel coordinates directly to points in the XY plane. There is no vertical component to this map, which results in apparent artifacts on the sides of objects within a scene (e.g. points along a vertical section of a wall in the local coordinate system will have equal [x,y] coordinates, and therefore map to the same pixel even though the z-position varies). That said, this type of map is very easy for mapping terrain properties (materials, textures, etc…​) onto a facetized surface generated from a height field.

The tile projection has the important property that it is geometry independent. This can be a useful feature for mapping nadir view imagery onto a terrain model with buildings, roads, etc.., effectively mapping the (top of the) entire scene at once. However, if there is a moving object within the scene such as a car, the geometry will move independently of the map (similar to an object being moved under a tablecloth). Moving objects should always use a geometry based projection, such as the UV projection described below.

In order to position the map within the scene, the user must provide an insertion point in the local coordinate system as well as the size of the pixels on the ground (an effective GSD, or ground sample distance, of the side of a pixel).

where the Insert Point is in Scene ENU coordinates and the GSD is in meters.

In contrast to the drape projection, the UV projection is geometry based. The columns of pixels in an image are represented on the continuous interval U in [0:1], corresponding to the full width of the image; similarly, the rows of pixels are represented on the continuous interval V in [0,1], corresponding to the height of the image. Every mapped facet in the image is required to have vertex texture coordinates in this UV space that maps that vertex to the image. The UV coordinates of a point internal to a facet is found by applying its barycentric coordinates to the vertex coordinates (i.e. it is a weighted sum of the values at the vertices).

A "bump" map is a common method used in the computer graphics community to introduce apparent surface height variation into smooth surface. The most common example in the literature is to impart the surface height variations of an orange or golf ball onto an otherwise smooth sphere. The technique employs an image map which is used to deflect the surface normal across the surface it is mapped to using gradients in the image. Because the deflected surface normal varies, the leaving radiance of the surface exhibits patterns that mimic what would happen if the surface height (and relative angle to the camera) were actually varying. However, a bump map does not actually displace the surface of the object (it does not "push in" the surface for each golf ball dimple).

The BumpMap1 demo features an "orange" (as in the fruit) bump map applied to a sphere geometry primitive. The first image below shows the resulting radiance image from the simulation. The shading across the "orange" varies because the bump map manipulates the well behaved normals across the sphere in a way that mimics the surface variations a "real" orange. The middle image is the "normal to view angle" truth image, which shows how the bump map is manipulating the surface normal. The right image shows the average Z intersection truth of each pixel, which highlights that the bump map does not change the location of the surface, but rather the normal at the surface.

The main parameter for the bump map is the scale parameter which manipulates how pixel digital count gradients are translated into surface normal deflections.

Normal mapping is a technique used to add the appearance of varying surface normals (eg. bumps and/or dents) to a geometric surface that lacks the fidelity required to show these features in the geometry itself. Normal maps differ from bump maps by explicitly encoding the relative normal into the R, G, and B channels of an image.

The NormalMap1 demo shows an example normal map setup. It showcases the difference between modeling surface topology with normal maps and modeling the same topology with facetized geometry.

The optional SCALE variable can be used to scale the normal map. The default value is 1.0 and a value of 0 will essentially disable the map.

A "decal" map is a type of geometry manipulation map that is fairly unique to DIRSIG (to the best of our knowledge). The idea behind it is that we often want to place fairly flat geometry onto other geometry in a way that the first object follows the topology of the second. For example, if our terrain geometry and our uv-mapped road geometry are coming from two different sources (e.g. a DEM and a procedural engine, respectively), it is almost impossible to get the facet edges to map up exactly so that the placed geometry always lies right on top of other.

If the two pieces of geometry are intended to match exactly (e.g. the placed geometry has no depth) then it may be advantageous to only use the underlying geometry and map the properties of the placed geometry onto it (see material and texture maps, below). However, if there is some depth (e.g. a curb) or it is easier to do all the mapping on the placed geometry itself (e.g. if the terrain would require too large a map to reach the desired quality), then the decal map provides a mechanism to generate the desired result.

The decal map works by modifying the position of the mapped geometry so that bottom is always just above the surface it is mapped to. Additionally, the object is rotated so that it aligns with the local tangential plane. The mapped geometry is defined in a GLIST (note that there is no direct ODB support though an ODB can be instanced in the glist) and, like the other maps, is assigned to geometry via a list of material ids. For example, the entry in the scene file should look like:

Note that the vertical translation of the object being mapped is ignored — the physical bottom surface is mapped to the underlying geometry.

The DecalMap1 demo shows an example of using this type of map to place a road on a sinusoidal "terrain".

The decal map is only intended for relatively flat geometry. The geometry that is mapped will not cast proper shadows, will not be seen at all angles, and will not maintain the rigidity of objects. For example, the following image shows a box of varying height being decal mapped onto the same sinusoidal surface; the top, shallow box behaves as desired, but with increasing height, the limitations of the map become more apparent.

The material map is the most commonly used map and associates a class of material (one of the entries in the material database) with geometry. This association is done through a digital count look-up table that matches pixel values to material IDs. Any supported grayscale or RGB image type can be used as input (only the red component is used in the case of RGB images for now).

A mixture map can be used to represent a surface that is a mixture of two or more distinct materials in the material database. The radiometry model computes the surface-leaving radiance for each material and then weights and combines them based on a set of mixing fractions.

The input to this map is a multi-band image file that contains N bands to define the fractions of N materials at each location. The image file must have the following properties:

The user must provide the list of material IDs that map to the different fraction "bands" in the image curve. In the example below, the image file mixture.img contains 3 bands, with each band containing the fraction of a specific material for a given location. The Band material ID assignments list indicates that bands 1, 2 and 3 (or 0, 1, 2) will map to material IDs 6, 7 and 8, respectively.

One of the primary mechanisms utilized in DIRSIG scenes to introduce variation within materials is a "texture map". However, the term "texture map" is generically used in the Computer Graphics community to describe the direct use of an RGB image to define RGB "reflectances" of an object. However, the DIRSIG model must be able to define the reflectances of surfaces at wavelengths outside of the visible RGB region. Furthermore, real materials have specific spectral correlations, where the variations at some wavelengths are correlated to variations at other wavelengths (either positively or negatively). Therefore, a method to introduce spatial and spectral variance was needed.

The "curve map" is available as an alternative to the classic DIRSIG texture map approach. Rather than relying on the statistics mechanism of the texture map approach to construct the curve indexes corresponding to digital counts (DCs) in the supplied map image, the curve map interprets the DC values in the map image as the index into the associated curve table. This places the user in charge of creating the curve index image that is used to define where spectral curves will be spatially mapped into the scene.

Although originally intended to introduce variation within a single material, this map can be used to spatially map curves that correspond to a variety of materials. For example, the spectral database might contain spectral curves for grass, asphalt, concrete, soil, etc. and the map will place those curves into the scene. In this usage, it collapses the conventional material map and texture map hierarchy. However, this unintended usage comes with a few limitations:

The CurveMap1 demo features a pair of curve map setups. One reproduces the variability one might observe in a patch of grass and the other acts more like a conventional material map by reproducing a Macbeth Color Checker target with a spectral library of curves for they 24 different materials featured in that target.

A "reflectance map" is an optical property that pulls the hemispherical reflectance (HR) from a spectral image file. Although the image file can be a single band or an RGB image, the primary purpose is to allow spectral image cubes (10s or 100s of bands) to be used. The reflectance data is expected to be an ENVI binary data and header file pair. The data should adhere to the following requirements:

This tool defines the spectral reflectance of a material in the scene. Like all material descriptions, there is no requirement that it use a specific spectral sampling. Hence, the reflectance map image could be sampled on 10 nm centers and other materials driven by spectral curves might be sampled on 1 nm centers. Furthermore, the sensor looking at a material assigned a reflectance map can use any spectral bands to image the material. Any wavelengths imaged outside the spectral range defined by the spectral reflectance image will be flat extrapolated (the closest value within the spectral range will be used).

The ReflectanceMap1 demo features a pair of image files that are used to define the reflectance for two different materials in the scene:

The RGB map feature provided by the RgbImage optical property is a variation on the reflectance map that pulls the hemispherical reflectance (HR) from a conventional 8-bits per channel (PNM, PNG, TIFF, GIF, etc.) image file.

The EarthMap1 demo features a 16200 x 8100 JPEG from the NASA Blue Marble site.

The temperature map associates a temperature (in degrees Celsius) with geometry. The input to this map is an ENVI image file with a single band. The value of the each pixel in that band is the temperature used.