1.1 What is ImageMapper ?

ImageMapper, by Surface Optics Corporation's (SOC), is a user-friendly software package for defining the material composition of image pixels based on their spectra. Knowing the material composition of a pixel and the related properties of those materials allows realistic sensor simulation across the electro-magnetic spectrum.

ImageMapper outputs Material Composition Maps (MCM). MCMs describe the mixture and relative proportion of materials within a pixel. This is a significant improvement in sensor simlation over the use of cartographic products such as VPF and Land-Use/Land Cover data, which assign a single material to every pixel.

ImageMapper is simple and easy to use. An analyst displays an image, provides the software with a set of reference or training pixels by pointing and clicking in the image and ImageMapper takes care of the rest. Based on the associations made by the analyst for a few basis pixels, the software determines the mixture and proportions of materials for every other pixel in the image. This is sometimes referred to as sub-pixel classification.

ImageMapper also includes

• A rendering engine for producing color, reflectance and radiance images based on the derived MCMs. This provides immediate feedback to the analyst for refinement, correction and quality assurance purposes. Renderings can also be used directly in sensor simulation systems.
• A set of region definition tools which add a spatial component to the classification
• A set of simple image processing utilities
• A set of basic materials with reflectance measurements from 0.4 to 14 microns.
• Additional tools, data and measurement services to support sensor simulations and trade studies.
  • A workbench for analyzing material properties data for contrast studies.
  • A full rendering package which includes a radiative transfer model.
  • Compatibility with SOC's materials database containing high quality measurements for 200 of the most common materials in the environment.
  • Material properties measurement services

1.2 Why Material Composition Maps?

In contrast to single 'thematic' layers which are typical in cartographic data, MCMs capture the sub-pixel mixturing of materials in a pixel. Just as polygons gave way to the use of textures (per pixel color) for greater visual detail and realism, thematic layers (which are often described as a mix of polygonal, linear and point features) will give way to MCMs which capture the per-pixel variation of materials in an image.

When rendered, thematic layers produce an image in which a 'tree stand' appears as a single color or radiance value over a defined region, leading to a 'map' like appearance. In contrast, MCMs contain material mixtures based on the original image on a per pixel basis. This produces a rich mixture of radiances when rendered that are directly correlated with the original image. Each pixel in a tree stand will be represented by a mixture of leaf, trunk and soil for instance, instead of a single value.

1.3 Potential Uses

ImageMapper provides the user with the ability to generate Material Composition Maps for input to a variety of applications.

• Visible, Radar, UV and Infrared Sensor Simulations
  
• Target Signature Analysis Codes
  
• Land Use / Land Cover Systems
  
• Operator Training, and
• Low observable design

1.4 Compatibility, Platforms and Image Formats Supported

Operating Systems:

ImageMapper is written in Java and runs on Windows 95/98/ME and XP/NT/2000. A version which operates under UNIX/Linux operating systems is also available.

Image Formats:

The following image formats are supported on input:

• SGI RGB
  
• JPEG
  
• GIF
• TIF & GeoTIF (with some exceptions)
  

ImageMapper also imports nearly any type of uncompressed image file format.

Output is in a simple non-proprietary format or convenient standard formats such as SGI-RGB or 'tif' image formats for import directly into a rendering engine. The material map format is 'open' and freely distributable.

1.5 Unique Features

The user does not have to be a material classification expert in order to set up and run ImageMapper. The software provides the user with:

• Ease of Use
• Flexibility on input, supporting various image formats
  
• Flexibility on output (SGI RGB, TIF and native)
  
• An intuitive, easy to use interface for selecting mappings
  
• Choice of classification algorithms
  
• Runs on a broad range of computing platforms (Java based)
• Tools to perform simple image processing tasks
• Region definition tools that add a spatial dimension to the classification
  
• Built-in infrared radiance and color simulation capabilities for:
  • Quality Assurance or
  • as direct input to a sensor simulator
• Fast execution speeds :
  • 25 seconds for a 512 by 512 image with 8 reference mappings on a 700MHz AMD with 256Megs of memory using SAM/MD classifier, 13 seconds using MD only
  • 7 seconds to generate a 512 by 512 color image
  • 7 seconds to generate a 512 by 512 radiance image from 0.4 to 0.5 microns
    

ImageMapper is composed of two main functional areas: the mapper/classifier and the simulator. The classifier uses reference mappings provided by an analyst and through a choice of algorithms, produces Material Composition Maps (MCMs). The simulator uses the resulting MCMs to render color, radiance and reflectance maps for map refinement and as input to real-time sensor simulators.

2.1 Image Mapping/Classification

Classifying imagery is a mix of science and art requiring ingenuity and often a great deal of time and effort. Most classifications take several iterations of mapping, rendering, evaluation and refinement before a satisfactory result is obtained. Because the largest amount of time and effort required in a classification is spent in the iterative process of classifying, rendering and evaluation, ImageMapper's philosophy is to give the analyst a set of tools to perform these tasks quickly and efficiently.

Classification : In ImageMapper, classification or the association of materials to spectra in an image is done by the best image processing system known to man, the human brain. The analyst defines a set of reference vectors, known as end-members, by associating a sampling of spectra in the image with materials.

ImageMapper takes these end-members and uses them to derive material composition on a per-pixel basis for the remaining pixels in the image. By using spectra directly from the image being classified, environmental and optical system differences can to a large degree be ignored. This assumes of course that the atmosphere is relatively constant across the image and that the image itself is not a mosaic of images taken under different environmental conditions.

There are a large number of classification algorithms available for spectral classification of imagery , supervised and unsupervised; k-means, parallelipiped, mahalanobis, spectral angle, maximum likelihood, iso-data, neural networks, and the list goes on. No one algorithm is superior under all circumstances and each has its inherent strengths and weaknesses. Attempts to make associations between observed spectra in a pixel and a reference set of materials automatically, without human intervention, have for the most part failed. This is due primarily to uncertainties in the environment at the time of data acquisition and the limited amount of reference spectra available for comparison. One of the primary contributors to uncertainty in classification is the environment at the time the imagery was taken; atmospheric absorption (an overcast or clear day) can have a profound effect on the spectral signature of a material as seen by a sensor. Classifiers work by selecting the material whose spectra or color most closely resembles that in the pixel. If conditions such as haze change the spectra, the results of the classification will suffer.

Classification Algorithms

ImageMapper makes available to the analyst three algorithms for generating MCMs. Two of the simplest and perhaps 'best' classification algorithms available; minimum distance and spectral angle mapping, and a third, hybrid algorithm, Angle/Distance that takes advantage of the strengths of both and also overcomes some of their weaknesses.

Why were these algorithms chosen ? Because they are simple, easy to understand, fast, and give a sharp analyst the capability to better interpret and modify results. They were also chosen because each has strengths in one of the two main types of images analyst are likely to encounter, high resolution images taken at close range and satellite imagery of broad geographic backgrounds.

Minimum Distance classifiers (MD) measure the distance in 'n'-space (n=3 for RGB) between the end-points of reference vectors and the spectral vectors of the pixels in the image. In cartographic applications, the classifier looks for the reference material which it is closest to the target pixel in 'n'-space and assigns it to that material. In material mixturing, the three closests reference materials are chosen and their relative proportions in the pixel are weighted according to their distance. This algorithm works surprisingly well under many circumstances, particularly in cases where color and intensity are well defined and associated with a single material.

The drawback of this method is that pixels with intensity can overwhelm color and mislead the classifier into choosing the wrong materials, i.e. it is difficult to distinguish between bright objects if their colors are similar. This can result in an unphysical mixtures of materials, which might appear visually pleasing at first, but can have serious effects in other portions of the electromagnetic spectrum.Visually, it might not seem important that a bright metallic surface got mixed in with bright concrete, but when performing a radar simulation, it can be seriously misleading.

Minimum distance works very well in instances where an image has good color and intensity separation, or simply good color separation, because similar intensities will lie along separate axes.. An image of a building front is a good example. Colors are separted in color space, are normally associated with a single material, and the minimum distance classifier works well.

Spectral Angle Mapping (SAM) measures the angle (dot product) between reference spectra and the spectra of the pixel in question. As with the minimum distance classifier, cartographic applications finds the reference vector with the smallest angle between it and the pixel in questions and assigns that material to the pixel. In material mixturing, the three closests reference materials are picked and their relative proportions in the pixel are weighted according to their angular separation. SAM works well at identifying colors regardless of intensity, which makes it an excellent choice for shadow removal. This is both a strength and a weakness.

Using SAM, materials in shadow, whose vector points in the same direction as their non-shadowed counterparts will be accurately classified as the same material. This can be a problem in something like a building front where shades of yellow may be entirely different materials. Something that a minimum distance classifier, which utilizes intensity, could easily separate..

SAM works well on large geographic satellite imagery where this algorithm can compensate for haze and shadows. Color separations are also more subtle and SAM will be more accurate than minimum distance algorithms at picking out differences.

SAM/MD classification is a hybrid approach, using the strengths of spectral angle mapping combined with the strengths a of minimum distance classifier. It works by first sorting the spectra by color and then uses intensity to perform the final classification. This insures that color matching is the primary criteria for end-member selection, negating the shortcomings of a pure minimum distance classification and then uses brightness information into the classification to overcome SAM's inability to differentiate on intensity.

SAM/MD's primary drawback is that it is more computationally intensive either SAM or MD. SAM/MD requires three additional MD calculations per pixel.At worst case, when only three reference spectra are chosen, it can be twice as slow. In cases of more than three spectra, the relative slow down is not as significant.

For further information on the minimum distance and spectal angle algorithms check the internet, There is a large body of literature available devoted to classification, including SAM and the minimum distance classifier.

2.2 Simulation:

ImageMapper's simulation capabilites are physics based, fast and accurate. The software simulation engine uses an MCM and has the ability to produce three different types of images; color, integrated infrared radiance and band averaged reflectance.

• Color: ImageMapper produces a color image of the MCM using the tristimulus function. A color image generated from an MCM can be a valuable source of feedback for refining mappings. It provides the analyst a means of comparing a derived image, based on their mappings, with the original. Pixels in the simulated image are assigned colors based on a set of measured spectral databases and the material mixture map.
• Reflectance: Reflectance images represent the average diffuse reflectance (and thus emissivity) of pixels over a wavelength region ... a value which for all intents and purposes does not change with lighting conditions or time of day. This can be a useful input to sensor simulators that do dynamic simulation. Instead of calculating these values at rendering time, these values can be stored as an 'alpha' plane in a texture to modulate the intensity of a unity reflector or the emittance of a unity emmiter, thereby removing some of the computational burden from the simulator and giving nothing up in terms of fidelity.
• Radiance: ImageMapper, when purchased with the radiance engine, is capable of producing radiance images in the visible through far Infrared regions of the spectrum for any observer/target geometry and time-of-day. The radiance engine incorporates the well validated LOWTRAN radiative environment model. In the thermal IR, the analyst supplies a temperature for each material and region. The software calculates the blackbody radiance, emissivity, path radiance and transmission to determine the pixel radiance. In the near future, ImageMapper will include a 3-node thermal model for terrain temperature prediction and support facet normal or elevation maps for solar illumination effects.

3.0 User's Guide

In practice, an ImageMapper session consist of 6 steps:

1. opening an image to map,
2. assigning a set of reference materials to colors by pointing and clicking on the image,
3. processing the mappings to produce material composition maps (MCMs),
4. rendering an image from the classification to check the quality of the classification,
5. refinement of the classification based on comparison between rendered and original images
6. rendering the desired output images (if desired), and
7. saving the final results.

3.1 Main Window

There are three areas in the main window, the material selection area (on the left), the central display/classification area in the center, the Navigation/Region/Zoom area (on the right).

The Material Selection area contains the Material Library , Material and Mappings panes. This area is used for selecting materials from the material libraries for mapping and for keeping track of reference mappings.

In the center Classification/Display work area is the Classification/Display pane, a '1X' display of the image or portion of the image to be classified is displayed. This is the area in which assignments are made. It is also the area where rendered images are displayed.

Directly below the Classification/Display pane is the Pixel Informer, a text-bar reporting cursor location and value of a pixel, 'RGB' or otherwise, depending on the type of image being displayed.

On the right, in the Navigation/Region/Zoom area are the Navigation pane, Region Tools, and Zoom pane.

The Navigation pane is used to navigate through large images, changing the focus of the Classification pane to be centered around where the analyst clicks in the sampled version shown in the Nav pane.

The Region Tools are used to define and manipulate regions.

The Zoom pane provides a magnified view of the area surrounding the cursor in the classification pane and can also be used to define end-members.

Regions can be used to partition and attribute pixels based on their spatial features. Regions are a powerful tool, providing a spatial element to a classification that can be used to;

• discriminate between spectrally confusing materials
• discriminate between urban and rural areas
• isolate homogeneous materials and remove noise
• provide a spatial level of thermal attribution which is often needed when the substrate of a material has a greater effect on its signature than the material itself.

Creating, selecting and removing regions is simple using the region tools.

There are seven menus on the menu bar: File, Edit, Render, Utilites, View, Display and Help. Below the menu items are shortcuts  to many of the often used functions.

The File Menu:

Open :

• Image : Open an Image for Mapping. TIF, GeoTIF, GIF, JPG, RGB and user-defined image formats are supported.
• Library: Open a material library and add it to the list of materials available for mapping. A list of additional libraries available from SOC are located in the 'additional libraries' folder in the ImageMapper directory. If you have purchased the additional libraries from Surface Optics Corporation, the libraries will be loaded upon selection.
• Mtf: Use a mapping developed from another image on the current image. For images that have been partitioned into more managable pieces for a rendering engine, this is an extremely useful feature, allowing the reuse of mappings developed for one portion of an image to be automatically loaded for use on another without having to re-enter the values.

Save Mappings: Saves the current mappings to the Map Text File.

Save Image: Saves a copy of the image under a different filename.

Close: Closes the current classification session.

Recent Files: Lists the last four image files processed for quick access.

Exit: Exit the software.

Preferences Window: Sets display and processing preferences.

• Mapping Delta refers to the delta in counts around a reference mapping's color in color-distance space that will be highlighted upon selection in the classification window. The default is 10, meaning any pixel within a color radius of 10 around the color selected will be highlighted. In images with low contrast, this value should be set to a lower number.
• True Color provides the option to use either the reference material's true color in overlaying pixels with the color associated with a material OR using a false color, for example bright yellow for water. False colors are sometimes easier for visualizing mappings. The default is to use True Colors. Note: False colors are assigned through the material library file, if a false color is not available (0,0,0), then true color is used in its place.
• Help Browser: In the event that the html browser is not in the default location, that location can be set by pushing on the locate browser button.
• Output Format: ImageMapper can output derived files in one of two formats; SGI RGB or Tiff. Select which depending upon your application.
• MIDIS Hardware: Selecting this button allows the software to use the MIDIS vector processor to perform calculations. If your computer is not equipped with the MIDIS processor, this selection will be grayed out. For more information on SOC's MIDIS vector processing engine see http://www.surfaceoptics.com.

Accept/Cancel: Accept or cancel the changes.

Header Window:

The header window allows editing of parameters that describe to the software the format and size of the image currently being displayed. It also pops up in the event that the image is not one of the standard types and the format is not recognized.

Cols - The number of columns or samples in the image.
Rows - The number of rows in the image.
Bands - The number of bands in the image, including and alpha channels.
File Offset - The number of bytes in the file header.
Bands Offset - The number of bytes before a band of information.
Line Offset - The number of bytes per line to offset.
ByteOrder - Wintel (little endian), IEEE (big endian)
DataType  - byte, short, integer or float
Interleave  - BSQ (Band Sequential), BIL (Band interleave by Line), BIP (Band interleaved by pixel)

Accept/Cancel: Accept or cancel the changes.

RGB : Create a color image from the classification.

There are no input parameters for creating a color image. Colors already associated with the materials mapped by the user are combined to produce the output color image. The user does have a choice, through the File/Preference menu, of using the true colors derived from the material spectra or 'false' colors, which are specified 'rgb' values found in the material library text file.

Reflectance : Create a band averaged reflectance image for a specific bandpass from the classification.

When a reflectance image is to be rendered certain information must be supplied in order to perform the calculation. For reflectivity, this is the spectral band over which the reflectance calculations are made. Reflectivities can be convolved with a Sensor Response Function through a sensor response file selected using the Sensor Response Function button. Rendering conditions and thermal parameters are not an issue in reflectance calculations. The resulting reflectance values in the rendered image are band-averaged.

Radiance : Create a radiance image

Similar to creating a reflectance image, when a radiance image is rendered certain information must be supplied to the renderer to perform the simulation; the spectral band over which the radiance calculations are to be made, a description of the radiative environment and temperatures for the materials and regions in the image. Radiances can also be convolved with a sensor response function and the radiative environment changed to reflect conditions during the simulation.

Spatial : In addition to spectral characteristics,  the spatial characteristics of pixels surrounding a pixel, its texture, can play a useful role in the classification of an image. One simple example of this is separating a green car and a green tree by examining the uniformity of pixels in the surrounding area. ImageMapper provides three methods for quantifying the spatial structure surrounding a pixel and producing an image layer from these measures

.

They are Haar Wavelet decomposition, Markov Random Fields and Grey-level co-occurence matrices. When these analyst 'renders' a spatial map, the software will produce an image of the same size as the original image containing the spatial measure for the algorithm selected which can be used in classification.

Note: At this time the inclusion of spatial layers can produce artifacts in the classified image. Examine the results of any spatial/spectral classification for errors.

Utilities Menu:

Through the Utilities Menu you can gain access to a number of utilites for image processing and manipulating data.

Clean MTF : Cleaning the MTF removes all unused materials from the MTF library.

Image Processing :

A set of simple image processing functions are available to aid in the classification. These fuctions are primarily for convenience and ease of use.

Reduce : For very large images, Reduce creates a temporary image of reduced size for faster panning and classification. Later, using the Open Mapping menu item, apply the mapping developed on the low-resolution version of an image to process the higher resolution image.

Extract : Extracts portions of an image from a larger image. Similar to reduce, only working on a portion of an image results in faster panning and classification. For very large images, mapping a small area that is characteristic of the entire image, might be sufficient. Later, using the Open Mapping menu item, apply the mapping developed on the extracted version of the image to process the entire image.

Statistics : Generate and view statistics for the image being displayed.

Convert : Convert images from one data type to another, e.g. integer to byte.

Combine : Used to combine single band images into RGB images for classification. Some remote sensing data comes as a series of single band images.

Spectral Analyzer :

The spectral analyzer is a powerful tool for interrogating spectral data to determine contrasts and features which might help in target detection. With the Spectral Analyzer it is possible to compare material reflectivities or radiances spectrally. Make plots of reflectances, source radiances and apparent radiances, take differences, determine contrasts and make potential design decisions.

Atmospheric Tool :

The atmospheric tool provides a graphical user interface to the Lowtran radiative transfer code used in ImageMapper to characterizes the radiative environment for rendering. The list of options available in Lowtran are too numerous to list here. For more information on running Lowtran and the meaning of the input parameters see the Lowtran User's Guide.

 Batch Processing :

Use batch processing to process multiple files. The processing will run as a background process, however ImageMapper will need to remain open in order for the processing to complete. However, during batch processing the analyst can minimize the application.

FILE - Opens an analyst batch file for processing.
Add - Adds a selected file for batch processing.
Remove - Removes the selected file from batch processing.
Go - Starts Batch Processing
Cancel - Cancels without processing.
Rendering Options:
• Select which output files will be rendered (color, reflectance, radiance)
Processing Options:
• Overwrite Existing Files - If selected, all current files will be overwritten.
• Algorithm - Select the classification algorithm which will be used in processing. Applies to all files.
Radiance and Reflectance Parameters
• Wavelength - Set the beginning and ending wavelengths for the reflectance and radiance calculations. These values can not exceed the bounds of the current atmospheric database.
• Gain and Offset - If you wish to scale all your radiance images to the same Global Scaling; 1) Select the global scaling box and 2) Set the values.
Show Dialog - Brings up a reporting window that shows the progress of the processing.

Once you have selected 'Go' to start the processing, the software will prompt you for a default MTF filename. All images which have not been classified directly and do not have their own MTF file, will use the mapping information in the default MTF filename to perform the classification. There are two important aspects to remember when using the default MTF file; 1) region definitions are ignored and 2) the only mappings that will apply are those made for the 'base' region in the default MTF file.

View Menu:

Through the view menu the analyst can switch between the original image and derived images; Color, Reflectance, Radiance, Wavelet (both decomposition and recomposition), Markov, GLCM and the Material Composition Map (MCM) to view the results of the classification.

The Image Info menu item brings up the Image Information window which describes the characteristics of the current image being display.

Shortcut buttons are also available on the Main Panel that correspond to the view menu selections.

Image Info :

Provides a text description of the input image and scaling factors for radiance images created.

Display Menu:

Through the display menu, the analyst can select the bands of the image that are being displayed. The default assignments for color images are 1=Red, 2=Blue and 3=Green. You can also select a single band, panchromatic display if desired.

Pressing on the Select Bands menu item brings up the Select Bands to Display dialog box. Select the bands that are used for the Red, Green and Blue channels in the image or alternatively, display a single band panchromatic image by selecting the band to be displayed in the Red/BW pull down and then select Panchromatic.

Help Menu:

Opens this document. ImageMapper uses a web browser for displaying documentation and its default is set to look for Internet Explorer on a Windows 95/98/ME/XP machine. If the browser is not found, ImageMapper will ask you to specify where the browser is on your machine using the File/Preferences menu.

 

3.3 Window & Tools

The main window is divided into 4areas, left to right: the Material Selection area, the main Classification/Display area, and processing panes.

3.3.1 The Mapping Pane

The Material Selection area contains four elements; the Mappings pane, Mapping Tools, the Material Library pane and the Material pane.

Mappings Pane : shows existing material assignments. A color, not a pixel, in the image is associated with a material through a mapping. Each mappings contains the three band value, in most cases RGB, of the color assigned to a material and a description of the material itself. For example, in the window above, the entry:

R:031 G:028 B:021 Meadow Grass

means that pixels with an RGB triplet of 031:028:021 in the image are assigned to the material meadow grass.

Each time a new mapping is performed, a new mapping entry will be added to the list. In the example above, three RGB values have been mapped to three different materials. It is not necessary to have unique materials for each mapping, however, colors must be unique. If you map a second material to a color that has already been mapped, a prompted will appear to ask whether or not to replace the existing mapping with the new one. If the answer is yes, the mapping will be replaced.

Mapping Tools:

Assign Mapping Button:

The assign mapping button becomes active when the analyst right-clicks in the image. This freezes the display and fixes the RGB value to the pixel the cursor was currently on. The analyst can then select the appropriate material from the materials pallete and by clicking on this button, assign the selected material to the RGB value.

Undo : Undoes last mapping.

Highlight : Highlights the current mapping in the image. Pixels turn red in the image that have the same RGB value and any pixels that are within the Mapping Delta. It also sets the material and material library windows to their appropriate values and draws a small red box over the location of the pixel that served as the reference.

Trash/Delete : Removes the selected mapping.

Clear : Clears all mappings.

Material Library Pane:

The Material Library pane contains a listing of the material libraries which are available to the analyst for classification. At start-up, the libraries listed in the LibraryList.txt file located in the ImageMapper home directory are loaded. To add to this list, simply edit the LibraryList.txt file. To load additional libraries while working on a classification, go to the File menu and click on Open Library. Library files have a suffix of '.lib'.

TIP: Loading material libraries at start-up is time consuming, so don't add libraries to the default file unless you plan on using them in your classification. Instead, just add them as necessary using the Open Library command located in the File menu..

In addition to the default libraries that are loaded when ImageMapper initializes, another library is loaded whenever a new or existing image is loaded. This library is called the MTF library and contains all the materials used in the classificaiton.

Clicking on any of the libraries in the Library pane will display the list of materials associated with that library. These materials are then available for use in mapping.

Materials Pane:

The Materials Pane lists the materials contained in the library highlighted in the Material Library pane. It is in the Material pane that the analyst selects the material which will be associated with a pixel or color in the image. This can be done in one of two ways: 1) By selecting a material first, then assigning it to a color by clicking on the appropriate pixel in the image Classification pane OR 2) by using the middle button on the mouse and clicking in the image on the pixel of interest and then selecting a material and assigning it using the Mapping Tools Assign button.

Using the first method, if the analyst wants to map the material water to a body of water in the image, then the Library containing the material water is first selected. A list of the materials in this library are then shown in the Materials pane; including water. Clicking on water to set the material, then a pixel in the water body in the image, results in a mapping.

Using the second method. A pixel in the water body is selected first using the middle mouse button, then the libarary and material associated with water are selected, followed by pressing the Assign button in the mappings tools.

Using either method, the mapping appears immeadiately in the Mappings pane, and pixels with the same RGB value or within the Mapping Delta of that value will change in color to the color of water as shown on the material label just above the Materials pane. This color can either be set to the true color of the material itself or its false color equivalent, depending on how the True Color flag is set in the preferences menu.

3.3.2 The Classifier Pane

The Classifier window is where mappings are performed. The Classifier window contains a 1X display of all or a portion of an image.

If the entire image can not be displayed in the classifier window, the arrows surrounding the window can be used to move about the image. The focus can also be changed through the Navigation window.

Below the classification pane is a text field which shows the location of the cursor in the image and the RGB, radiance or reflectance value respective of what type of image is being displayed.Reflectances are in percentages and radiances are in units of watts/cm**2/steriadian.

Mouse Actions:

Left - click results in a mapping between the pixel selected and the current material.
Center - Freezes the color value and allows selection of material.
Right - closes a region when in region definition mode.
Shift-Left - Freezes the zoom window allowing classification and region definition to occur in the zoom window

3.3.3 Navigation/Region/Zoom Pane

The Navigation section contians four tools:

• Navigation Window
• Region Tools
• Zoom Window
• Display Tools

The Navigation Window displays a sampled version of the entire image and allows navigation of very large images. By clicking on a point in the navigation image or clicking and dragging on the red box, the center of the image displayed in the classifier window is changed.

Region Tools:  Regions give the analyst the flexibility to derive a custom set of mappings and temperatures for defined areas on the image. Each region has its own sets of mappings and attributes which are processed separately from other regions. For example, if a region is defined around a car, the analyst can set the color green to correspond with green paint in that region and allow  green in surrounding areas to correspond with grass. The temperature of the car can also be different.

There are six tools in the region tool window:

• New - Press to begin defining a new region.
• Close - Press to close a region or right-click in the classification window
• Cancel - Press to cancel the current region definition
• Region Select - Press to select an existing region with the mouse or  from a list.
• Remove- Press to delete the selected region
• Undo - Press to undo the last point specified during region definition

If Snap is used. Points on the region will snap to the closest edge that lies within the tolerance value selected.

The Zoom Window : The zoom window displays a zoomed version of an area centered around the cursor in the classification window. In the zoom window, both mapping and region definition can be performed. To freeze the zoom window, press Shift-Left in the classification pane. The zoom window will unfreeze the next time the classification window is entered.

Display Tools : The display tools allow stretching and brightness adjustments to be made to the image being displayed.

Pressing on the M (classification), Z (zoom) or N (navigation) buttons results in the image being stretched across the image values displayed in the corresponding window. This can be helpful in instances where the analyst wishes to stretch the image displayed out over a smaller dynamic range. The R, G and B buttons and slider allow the analyst to adjust the brightness in the display of any or all bands in the image.

3.3.4 The Display Toolbar

Display Select : Shortcuts to the View menu. These buttons allow quick toggling between the original and derived or rendered images.

Show Mappings : allows the user to toggle the display between showing and not showing mappings and region graphics.

Blink : Toggles between the current image and the last image displayed. This can be useful for comparing the results of a classification with the original, or any other  image. When pressed, the last image is displayed, when released, the current image is displayed.

Process: The process button forces a new classification to be performed. This button is red when new mappings or regions have been added, it is green otherwise.  

4.1 Output Format

Binary Image File Format: With the exception of the mtf file, all other outputs from the ImageMapper software are binary image files. This includes rendered images, such as Radiance '.rad', Reflectance '.rfl' and Color '.clr' files  and also the mcm file.

Multi-band image data is normally stored in one of three ways; Band-Interleaved-by-Pixel (BIP), Band-Interleaved-by-Line (BIL) or Band-Sequential (BSQ). Normally there is also a fixed header attached to the beginning of the image which describes the format and any other information associated with the image.

ImageMapper outputs images in one of three formats; Native, SGI's RGB and Adobe Systems' 'tif' format. In general, the native format is only used to produce uncompressed versions of compressed files and carries the suffix, '.imp'. The rest of the files are either in RGB or tif formats. The choice of which of the two formats is used is left up to the analyst and can be set through the preferences menu.

Binary Image File Formats:

• Native format images consist of byte values in a BIP format with a '0' byte header (the image format information is stored in the '.hdr' files).
• RGB images are inverted relative to the way they are viewed, the upper left corner of the image occupying the lower left corner of the data on the disk. Keeping the output in RGB format allows rendered images to be used directly in rendering engines software. RGB image files consist of a 512 byte header, followed by the image data in a BSQ format. For a full description of SGI's RGB format visit SGI's website.
• TIF format files have an 8 byte header for a single band image (radiance and reflectance) and 14 byte header for color images. The TIF format document is available from Adobe Systems via the world wide web.

Radiance and reflectance values, which are inherently single-band 'float or real' numbers, are scaled and stored in  unsigned byte images, each byte representing a scaled value. The scaling for an image can be found using the Image Info selection in the View menu.

Color images are three-band byte images with three bytes per pixel, stored in either RGB or TIF format.

Material Classification Maps (MCMs) are five, six or seven-band images with five, six or seven bytes per pixel, also stored in either RGB or TIF format. The first five bands contain the materials; material1, material2, material3, and the weights or mixtures of the first two materials within the pixel, the third value being 1.0 - the sum of the other two. The optional 6th and 7th bands contain the alpha channel from the original image (if one existed) and the region map (if regions were defined).

Mat1 Mat2 Mat3 %1 %2 Alpha Region ...... Mat1 Mat2 Mat3 %1 %2 Alpha Region

Materials are stored as indices which are associated with their appearance in the map text file (.mtf). The value '0' corresponds to the material listed first in the map text file, 1 to the second material, and so on.

The map text file is the 'key' to the classification, do not remove it or modify it unless you completely understand the consequences of doing so. In some cases, modifying the map text file can save time and allow manipulation of the mapping without invoking the ImageMapper tool, but use this option with caution.

Map Text File (.mtf) Format (ASCII):

Map Text Files (.mtf) are self documenting, with parameter names listed immeadiately after the value. An example MTF file is shown below.

• Version 2.0
• C:\Documents and Settings\User\Desktop\Data\LAV120wheelDes.rgb {Input Image}
• 256 256 3 {Columns, Rows, Bands}
• 4 {Number of Libraries}
• MTF File {Library #0}
• .\\Natural_THR.lib {Library #1}
• .\\ManMade_THR.lib {Library #2}
• .\\Reference_THR.lib {Library #3}
• 2 {Number of Regions}
• 'Base' {Region Name}
• 0.0 {Temperature}
• 4 {Number Of Vertices}
• -1 -1 {Vertex #0}
• 258 -1 {Vertex #1}
• 258 258 {Vertex #2}
• -1 258 {Vertex #3}
• 3 {Number of Mappings in Region}
• 169 {Band 1}
• 154 {Band 2}
• 115 {Band 3}
• 0 0 0 0 {Library, Material, X, Y}
• 12 {Band 1}
• 11 {Band 2}
• 9 {Band 3}
• 0 1 0 0 {Library, Material, X, Y}
• 33 {Band 1}
• 30 {Band 2}
• 22 {Band 3}
• 0 2 0 0 {Library, Material, X, Y}
• 'lugnut' {Region Name}
• 395.0 {Temperature}
• 5 {Number Of Vertices}
• 134 192 {Vertex #0}
• 168 164 {Vertex #1}
• 140 128 {Vertex #2}
• 106 132 {Vertex #3}
• 91 156 {Vertex #4}
• 3 {Number of Mappings in Region}
• 169 {Band 1}
• 154 {Band 2}
• 115 {Band 3}
• 0 0 0 0 {Library, Material, X, Y}
• 12 {Band 1}
• 11 {Band 2}
• 9 {Band 3}
• 0 1 0 0 {Library, Material, X, Y}
• 33 {Band 1}
• 30 {Band 2}
• 22 {Band 3}
• 0 2 0 0 {Library, Material, X, Y}
• 3 {Materials}
• ' Construction Asphalt' {Material Name}
• 0 0 {Lib, Mat}
• 300.0 {Temperature}
• 491 {Number of reflectance values}
• 0.42 4.0837
• 0.422 4.6136
• 0.424 4.7327
• 0.426 4.8886
• 0.428 4.8321
•          .
•          .

4.2 Input File Formats

4.2.1 Material Properties File Format (ASCII) Material Properties files contain spectral information and other quantities that characterize the material for sensor simulation.

Lines 1->6 : Comment Lines (to be used at a later date)

Line 7: Number of Spectral Reflectance Points (numberOfPoints)

Lines 8 -> (7+numberOfPoints) : Wavelength (in microns), Reflectance <CR>

4.2.2 Material Library File Format (ASCII) Material Library Files are basically lists of the materials in a particular library. ImageMapper reads these files and loads the material properties files in the material library file into the tool.

Line 1 : Library Name

Line 2: Number of libraries (numberOfLibraries)

Line 3 -> (2+numberOfLibraries) : fileType, 'Material Proporties Filename', R, G, B

** R, G, B int values between 0 and 255 which are the 'false' colors for the material **

4.2.3 Preferences File (Pref.txt) Contains a number of preferences set by the user that are loaded at startup into ImageMapper.

Line 1: Number of Libraries to load upon initialization

Lines 2 -> (1+intiLibraries) : 'Library Filenames'

Next Line : Mapping Delta

Next Line: Location of browser

Next Line: Coloring Scheme to Use (0=falseColor, 1=trueColor)

Next Line: Location of last file processed