Reference Manual for the TNT products V6.50
Previous Page    Next Page    Bottom of Page    Table Of Contents    Index    Search   
344.2.4

Orthorectification Mode

Orthorectification corrects a digital image, such as a scanned aerial photograph. The process uses the image's orientation statistics and either a DEM object that covers part or all of the same area as the image, anchorpoints (control points), or an average elevation to remove distortion from the image. You can orthorectify a single input image, multiple images, and/or vector and CAD objects that are associated with the image(s) during a single run of the Orthorectification process. The resulting orthoimage is comparable to a map in that it does not have the scale, tilt, and relief distortions typical of unprocessed, photographic images. That is, your viewpoint of each pixel in the orthoimage is orthogonal (perpendicular) to the earth.

You can use orthoimages as base maps in layouts or in the generation of vector or CAD objects that trace road, hydrology, or other features. You can also accurately measure distances, areas, and angles directly from an orthoimage, which you can not do with the unprocessed image.

Choose Orthorectification from the Mode option button menu located below the menu bar of the Digital Photogrammetric Modeling window. Next click the Left Image button to open the Select Object window (you are not actually limited to a left image but can use any aerial photograph). Select the image that you want to orthorectify and then click the OK button at the bottom of the Select Object window. Another Select Object window opens with the message "Select ground control point object." Choose the control point object that you want to use in the orthorectification process. Some images have only one control point object, while others have multiple control point objects, each using a different projection. Click the OK button after you select the object to close the Select Object window and return you to the Digital Photogrammetric Modeling window.

ii/iprsmorp.gif

Orthorectification Mode, Parameters

Now that you have an image and control point set selected, you need to choose the parameters that the orthorectification process will use. The Parameters tabbed panel is selected by default when you choose the Orthorectification Mode. You need to first choose the Process Mode as it determines whether you need to select a DEM object for use in the Orthorectification process. The available choices are Pixel by Pixel, Anchorpoints, and Transformation. Your choice of Process Mode depends upon the availability of a DEM for the area covered by the image and/or the relative flatness of the area. Pixel by Pixel is the best processing mode for areas with irregular terrain, such as mountainous regions. However if you choose Pixel by Pixel from the Processing Mode option menu, you must have a DEM to run the process. You can use an existing DEM or a DEM that you created in Prospective Projection / DEM Extraction mode. To select a DEM object for use in the process, click the DEM button on the Parameters tabbed panel. The Select Object window opens. Choose the DEM object that contains the area covered by the input image you wish to rectify. When you click the OK button at the bottom of the window, the Select Object window opens so you can choose the appropriate control point object. After you choose the object, click the OK button to close the window and return to the Digital Photogrammetric Modeling window.

NOTE: The DEM and the input image do not have to be the same size. However, if the DEM is smaller than the input image, the output raster object from the Orthorectification process is clipped to the size of the DEM.

If you do not have a DEM of the area covered by the input image, you can still orthorectify the image using one of the other two Processing Modes. Anchorpoints and Transformation give the best results if the area covered by the image is relatively flat. However, you can also use Anchorpoints for areas that are not flat when you do not have a DEM object for the area covered by the input image. The Anchorpoints processing mode uses the elevation coordinates of the control points of the image in the orthorectification process. If your image covers an area that is not particularly flat, you should make sure that the control points cover the elevation range in the image. You can add additional control points in the Georeference process (Edit / Georeference) if they are needed to give a good representation of elevation variation. You can read how to do this in the Georeference chapter of the Edit volume.

NOTE: Widely dispersed control points give better results for all processing modes in the Orthorectification process than control points that occur in clusters or are concentrated in a single area of the image. This is especially important when you are using the Anchorpoints mode to generate an orthoimage because you lack a DEM object of the area.

The Transformation processing mode calculates an average elevation from the control points and uses the average to remove elevation angles and transform the image into a planar image. You may prefer to use this method even when you have access to a DEM. If your image covers a flat area (difference in minimum and maximum elevation is no more than 20 to 30 meters) and resolution of your image is 10 meters or more, the Transformation processing mode generates a satisfactory orthoimage.

Resampling is necessary during orthorectification because cell values of the rectified image must be fit into a new grid of rows and columns (the grid of the orthorectified image differs somewhat from that of the input image). You have three choices on the Resampling Method option menu-Nearest Neighbor, Bilinear Interpolation, and Cubic Convolution. The Nearest Neighbor method is the simplest in that it requires relatively little computation and does not alter the original pixel values of the image. This method assigns the value of the nearest input pixel to the output pixel. The Bilinear Interpolation method uses a grid of the four nearest pixels to calculate a distance-weighted average of the pixel values for each output pixel. This method alters the gray levels of the output image when compared to the input image. The Cubic Convolution method utilizes a block of 16 pixels (4x4) to calculate a value for each pixel in the output image. This method also alters the original, input gray values in the output image. The Cubic Convolution method avoids the sometimes jagged appearance that arises from the use of the Nearest Neighbor method and gives a somewhat sharper image than the Bilinear Interpolation method. However, it is computationally more intense and so requires more processing capacity and time. Your choice of resampling method is dependent on your input image and intended use of the orthorectified image.

The Orthoimage Cell Size is entered in the numeric field automatically when you select the image for the Orthorectification process. You also have two options that you can turn on, if you choose. Turn on the Rectify Co-Processing Objects Only check button if you want to orthorectify the co-processing objects but not the image you selected when you clicked the Left Image button (co-processing objects are discussed in a following paragraph). The Orientation Statistics of the selected image are used to orthorectify the co-processing objects. You can also choose to compress the output raster objects that are generated by the Orthorectification process. Simply turn on the Compress Output Rasters check button.

The Orthorectification process needs orientation statistics in order to remove distortion from an image. These statistics are computed from the control points and the camera parameters of the input image. You may see the message, "Camera parameters are undefined" after you click the Left Image button and select an image and control points object. When you receive this message, change the Prospective Projection mode from Orthorectification to Interior Orientation. Enter the camera parameters, click Save Camera Parameters, and return to the Orthorectification mode. Click the Compute button that is located above the Orientation Statistics panel, and the statistics are calculated. You may get a message window if the RMS (root mean square) errors are large. The message states "RMS errors for orientation process are too big. Accuracy of output orthoimage will be affected. Please check input parameters and georeference points. Do you want to continue?" Check the input camera parameters to make sure they were entered correctly. You may also need to redo some of the control points (the georeference point in the message). You can proceed with the Orthorectification process in spite of the RMS errors but the quality of the resultant orthoimage will be affected; that is, you will probably not have an orthogonal view of all pixels in the image.

Selection of co-processing objects occurs on the Co-Processing tabbed panel. Select the panel and then click the Co-Processing Objects button. The Select Object window opens. Choose one or more objects to use in the Orthorectification process. You can select raster, vector or CAD objects for co-processing. These objects must cover the same area as the image you initially selected when you clicked the Left Image button. For example, you have an input image that has associated vector or CAD objects that show roads, hydrology, or other features. You can orthorectify all the objects simultaneously. The Clear All button clears your selected co-processing objects. You can select another set of objects for orthorectification or have no co-processing objects when you run the process.

ii/iprsmorc.gif

Orthorectification Mode, Co-Processing Objects

You can view a report on the outcome of the Orthorectification process in Project File Maintenance. Open this process from the TNTmips main menu using the Support / Maintenance / Project File menu cascade. Select the orthoimage file and object. Double-click on the object name and then highlight the Report subobject. Click the Info icon on the toolbar near the bottom of the Project File Maintenance window. This opens the Object Information window where the report is displayed. You can scroll down through the report and see the input object names, the processing mode used, orientation angles, DEM cell size, and so forth. See the section entitled Managing Project Files, Objects, and Subobjects in the Support volume for more discussion of how to use Project File Maintenance.

ii/iprsmorr.gif

Report on orthorectification process

345

Resampling Raster Objects

The resampling processes interpolate cell values, or otherwise assign values to cells in a new raster object, and create a new raster object with different cell sizes and dimensions. Resampling is used to change the scale of an input raster, and it is also used in conjunction with geometric transformation models that change the internal geometry of a raster.

Both manual and automatic Resample processes currently employ three resampling methods. These resampling methods use interpolation and/or cell value assignment techniques to determine new cell values for an output raster whose dimensions are different than that of an input raster. The dimensions of an input raster may be increased or decreased by varying the raster size and/or the raster cell size. Raster resampling is a required process for preparing enhanced resolution illustrations where you use a higher resolution component for the fourth component in RGBI or RGBB display modes. For additional information on enhanced resolution illustrations refer to the section on Multiple Component Raster Display Controls in the Display volume.

The Resample processes, however, differ markedly in terms of purpose and the operations they perform. The Manual Input Raster Resampling process is primarily designed to numerically manipulate an input raster object in order to increase or decrease dimensions, and to reorient raster objects via rotation. The Manual Input Raster Resampling process does not and cannot be used to change the internal geometry of the input raster object.

The Automatic Raster Resampling process is primarily designed to rectify input raster objects by using geometric transformation models, or by making the raster conform to a specified map projection. Raster rectification, or rubbersheeting, corrects distorted rasters to produce geographically calibrated rasters in accordance with specified map projections. In the context of the Automatic Raster Resampling process, raster resampling methods are used in conjunction with geometric transformation models that change the internal geometry of a raster.

Geometric transformation (or warping) broadly refers to any process in which an image is stretched differentially so as to change its internal geometry. Redefining the spatial relationship between cells in a raster object can be achieved by any one of many transformations, such as changing a map projection, fitting a polynomial to a surface, least-squares movement of control points, and so on.

The Automatic Raster Resampling process requires each input raster object to be associated with an existing georeference control point subobject. A raster without a georeference control point subobject (or an implied georeference subobject) cannot be made to conform to another map projection, and each of the geometric transformation models require a minimum threshold of georeference control points. If your input raster is not associated with any valid georeference control point subobject, you may define one using the various tools and utilities of the Georeference process (Edit / Georeference).

The Priroda lens correction process resamples a raster object according to known lens characteristics in order to remove optical distortion and produce an image with suitable geometry for other processes, such as Stereo to DEM. The process is specifically designed for Russian Priroda satellite imagery, and has no application for other types of images. Moreover, this version of the process was developed for images of the KFA type. If you have other types of Priroda images, contact MicroImages, Inc. for a modified version of this process.

Generally, the Resample processes are used to prepare input raster objects that have the same cell size, dimensions, and orientation for other processes, such as Mosaic, Automatic Classification, or Feature Mapping. For example, you may be working with a series of air-video frames that vary slightly in orientation and cell size due to in-flight changes in the heading and altitude of the airplane. If the scenes contain enough reference information, you can use the Resample processes to give them a common cell size, to align them to a common orientation, and to rectify them.

Select Process / Raster / Resample from the main TNTmips menu to select from the following resampling options: Manual, Automatic, and Priroda Lens Correction. The Manual Input Raster Resampling process allows you to employ raster resampling methods to determine new cell values for output rasters with changed dimensions, and/or to change the orientation of an input raster via rotation. The Automatic Raster Resampling process employs raster resampling methods in conjunction with geometric transformation models that are used to rectify georeferenced input raster objects. The Priroda Lens Correction process resamples a raster object according to known lens characteristics in order to remove optical distortion and produce an image with suitable geometry for other processes, such as Stereo to DEM.

Understanding of the following terms is useful for comprehension of the raster resampling processes:
Resample
To resample is to interpolate cell values, or otherwise assign values to cells in a new raster object, and create a raster with larger or smaller cells and different dimensions. Resampling is used to change the scale of an input raster, and it is also used in conjunction with geometric transformation models that change the internal geometry of a raster.
Interpolation
Interpolation is a collective term for various techniques of determining an approximate value of a function at a point in the domain between given points at which the function values are known. Bilinear interpolation, for example, is based on the presumption that the value being sought lies on or near a straight line joining two known values.
Convolution
Convolution mathematically determines the value for new cells in an array of cells. Raster filtering, resampling, and other raster processes employ convolution. Convolution methods should not be applied to raster objects that contain discrete (or categorical) data. Convolution is only appropriate for continuous data.
Continuous Data
Data in a raster is continuous if it can be represented by a three-dimensional surface such that intermediate values can be derived with intermediate results. For example, an elevation raster has continuous data because an elevation of 400 and an elevation of 500 can be averaged to assign an intermediate elevation of 450.
Discrete Data (or Categorical Data)
Data in a raster is said to be discrete if it cannot be represented by a continuous surface because intermediate terms cannot be derived with meaningful results. With discrete data the representation of a variable may assume one of a finite set of values. For example, soil type data cannot be interpolated, since a soil type 14 and a soil type 15 cannot always be averaged to derive a soil type 14.5.
Geometric Transformation (or Rectification)
A process in which an image is stretched differentially so as to change its internal geometry is said to have undergone geometric transformation or rectification. A transformation specifically refers to the process of projecting an image from its plane onto another plane by translation, rotation, and/or scale change.
Georeference (or Geographic Calibration)
A georeference is information that relates raster cells or vector / CAD elements to a specified coordinate system or map projection. TNTmips maintains information needed to relate every raster cell or vector / CAD point to a specific coordinate system / map projection in subobjects associated with a parent object. Geographic calibration may be established when creating an object; for example, extracting an image map from a Landsat or SPOT satellite image. A georeference may also be established for an existing object by entering control points or by associating it with a previously calibrated object; for example, aligning a calibrated vector to a raster object. A TNTmips georeference subobject identifies a calibrated object's map projection.
3451

Manual Raster Resampling

The TNTmips Manual Raster Resampling process allows you to change the size and/or orientation of one or more raster objects by using linear and cubic resampling methods and by enabling you to rotate the raster(s) in either a clockwise or counter-clockwise direction. This process does not allow you to alter the internal geometry of the input raster. The Automatic Raster Resampling process, however, allows you alter the internal geometry of raster objects while providing most of the functionality provided by the Manual Input Raster Resampling process.

Select Process / Raster / Resample / Manual from the main TNTmips menu selection window to open the Raster Resampling (Manual Input) window. This window includes a Rasters button and an associated input rasters scrollable text field, and four panels where you may define relevant parameters for output rasters: Raster Size, Relative Cell Size, Input Limits, and Relative Cell Size. For example, rather than resampling the entire input raster to a new output raster, you can resample only a sub-portion, by entering ranges in the lines and columns text fields on the Input Limits panel.

ii/iprmwind.gif

Raster Resampling (Manual Input) window

Run initiates the resampling process after input raster objects, a resampling method, and requisite parameters are selected. You are prompted to select one or more output raster objects before processing begins. Exit ends the process by closing the Raster Resampling (Manual Input) window, and returns you to the main TNTmips menu selection window.

The Resample menu lists the resampling methods that are supported by the current release of TNTmips. A resampling method is selected using one of the following options: Nearest Neighbor, Bilinear Interpolation, or Cubic Convolution. (Nearest Neighbor is the initial default setting, but default status is subsequently assigned to the last selection made.) A description of these resampling methods is provided in a subsequent section entitled Raster Resampling Methods.

3451.1

Raster Size

After selecting one or more input rasters, specify the number of lines and columns in the output raster(s). The four text fields on this panel show the dimensions of both the input raster(s) and the output raster(s). The Input Lines and Columns text fields show the dimensions of the input raster(s), and these values cannot be changed. Initially, the Output Lines and Columns text field values default to those values shown in the Input text fields. You can enter new values in these Output text fields to specify dimensions for the resampled output raster(s).

NOTE: The values that you enter in the Output text fields are automatically changed whenever you change any given values on the Relative Cell Size panel or whenever you change the angle of rotation on the Rotation panel.

3451.2

Relative Cell Size

The Manual Input Raster Resampling process determines how much larger or smaller the output raster object(s) will be by the ratio of column and line widths between the input and the output. For example, if your input has a ten-meter cell size, and you want a five-meter cell size for the output, enter 5 in the input text fields and 10 in the output text fields on the Relative Cell Size panel. If you do not know the input raster cell size, but know that you want the output cell size to be twice as large, enter 1 in the input fields and 2 in the output fields. The ratio is what the process uses in the calculations, not the number themselves.

The four text fields on this panel show the relative cell sizes of both the input raster(s) and the output raster(s) in terms of line and column width. Given that the input raster object(s) serve as a bench mark, width values in the Input Lines and Columns text fields are initially shown as 1.000. The initial default width values in the Output Lines and Columns text fields are also the same as those values shown in the Input text fields. You can enter a new value in the four text fields to specify the relative cell size in the output raster.

Reconsider the initial example where your input has a 10-meter cell size, and you want a 5-meter cell size for the output. You can enter 0.500 in the Output Lines and Columns text fields or you can enter 2.000 in the Input Lines and Columns text fields. In either case, the relative cell size ratio between input raster and output raster is 2:1. A resampled output raster with a relative cell size one half of that of the input raster will have four times as many cells and twice the number of rows and columns.

If you specify a precise number of rows and columns for the output raster object, and if the resampled output raster is not rotated, you can enter values for the number of lines and columns in the Output text fields on the Raster Size panel. However, whenever you respecify the number of lines and columns in the output raster, the input to output relative cell size ratios are automatically adjusted accordingly. Likewise, if you adjust a value on the Relative Cell Size panel, values on the Raster Size panel are automatically adjusted accordingly.

For example, if you have a 488 x 361 input raster object and you want a 781 x 542 resampled output object, you can enter the values in the Output Lines (781) and Columns (542) text fields on the Raster Size panel. When you enter these values, relative cell width values on the Relative Cell Size panel are automatically adjusted as follows: Input Lines 1.000, Output Lines 0.625, Input Columns 1.000, and Output Columns 0.666. Notice that the column and line scaling factors (or relative input to output cell size width ratios) in this example are slightly different (0.625 not equal to 0.666).

IMPORTANT: Aspect distortion will occur in the resampled output raster(s) whenever the column and line cell size scaling factors are not the same. To eliminate aspect distortion, you can use the same scaling factor for relative column and line cell widths.

Reconsider the previous example where you have a 488 x 361 input raster and you want a 781 x 542 resampled raster object. Given that this resampled raster object will have some aspect distortion, you might consider eliminating this problem by adjusting values on the Relative Cell Size panel. If you changed the Output relative column width value from 0.666 to 0.625, for example, the column and line cell size scaling factors would both be the same. Making this adjustment, however, would yield a 781 x 578 resampled output raster, rather than a 781 x 542 output raster.

3451.3

Input Limits

The Manual Input Raster Resampling process allows you to resample only a portion of the input raster object(s). You may select a specific range of lines and columns to resample on the Input Limits panel. Minimum range values are shown by the two text fields on the right-hand side, while maximum range values are shown by the two text fields on the left-hand side. The default entries in all four text fields indicate that all lines and columns in the input raster(s) are currently selected for resampling. Enter new values in the two top text fields to specify a limited range of lines to resample, and enter new values in the two bottom text fields to specify a limited range of columns to resample.

3451.4

Rotation

The Manual Input Raster Resampling process also allows you to rotate input rasters. The Angle text field on the Rotation panel is initially set to 0.00000, presuming that you will resample input rasters without necessarily changing raster orientations. The Manual Input Raster Resampling process allows you to change the orientation of input rasters without requiring adjustments on the Raster Size and Relative Cell Size panels.

In addition to the Angle text field, the Rotation panel includes a units option menu (adjacent to the text field) and a pair of Direction radio buttons. To specify a different orientation for your output raster object(s), enter the angle of rotation in the text field after selecting the desired units of measurement. The default setting on the units option menu is degrees, but you can also specify the angle of rotation in radians, milliradians, or gradians. Toggle either the CW or CCW radio button to designate the direction of rotation as clockwise or counterclockwise, respectively. The Manual Input Raster Resampling process automatically calculates the necessary dimensions of the output raster object(s) whenever you rotate the input raster(s).

IMPORTANT Whenever you rotate an input raster by an angle that is not equal to a multiple of 90 degrees, the output raster will have wedges of null cells along the sides. For example, an output raster object with an equal number of lines and columns if rotated 45 degrees will appear like a diamond placed on a square black (null) background after processing.

3451.5

Default or User-Defined Null Values

Use the Null Value toggle buttons to choose either the Default value or a User-Defined value. The text field becomes active if you toggle on the User-Defined option and allows you to enter a value for value of null cells. If the Default option is selected, the null value text field remains inactive.

3451.6

Create Pyramid Tiers for Manual Resampling Output

You have the option to use or not use the pyramiding process to display output rasters by toggling on or off the Create Pyramid Tiers toggle button. The pyramiding process can be used to speed up display times for some raster objects. Sampling raster objects for display at zoom factors less than one can consume excessive display time, and as the raster size increases, more time is required. Pyramiding creates sampled rasters and saves them as subobjects of the original raster object. These subobjects require less disk space and display time than the original raster object. (For information on the pyramiding process, see the section entitled Raster Pyramiding in this volume.)


See also: Pyramiding Rasters

3451.7

Raster Resampling Methods

Select a resampling method from the Resample By menu after you have selected the input raster object(s). Options on the Resample menu include: Nearest Neighbor, Bilinear Interpolation, and Cubic Convolution.

The accompanying illustration presents four rasters, one showing an input raster object and one each showing the result of the three resampling methods. The first contains a 10 x 10 block from the upper left-hand corner of the Crow Butte red TM data set. The second is the Nearest Neighbor resampled version with twice the cell resolution per axis (half the cell size). The third is the Bilinear Interpolation resampled version with twice the cell resolution per axis. The fourth is the object created using the Cubic Convolution resampling method with twice the cell resolution per axis.

ii/cimanres.gif

Manual Resampling example results

Notice that the nearest neighbor resampling, when using an integer resampling ratio, results in a visually identical raster, while the others visually smooth the raster, interpolating new details into the raster. The new details may not be accurate, but they are likely closer to reality than is the stair-step appearance produced by the nearest neighbor method. Remember that only Nearest Neighbor is a meaningful resampling technique for rasters with discrete data types such as soil or crop categories.

3451.7.1

Nearest Neighbor Method

The Nearest Neighbor resampling method selects and locates the center of each cell in the output raster object. An equivalent location is found in the input raster for each output cell. For example, if the selected output raster cell's center is in the center of the output raster, that same relative location is noted in the input raster. The method computes the distance between the center location in the input raster object and the four nearest cells contained in the input raster and determines which of those four cells is located nearest to the center location. The process then assigns the value of the input raster cell located in closest proximity to the center location, to the output raster cell. Therefore, the input of one input cell may be assigned to more than one output cell. Conversely, some input cells may not be transferred at all to the output raster. These under sampling and over sampling situations occur when the cell sizes of the input and output raster are different. Under sampling and over sampling distorts the appearance of features in output raster objects from those in an input raster object.

Nearest Neighbor resampling is the only method suitable for use with discrete input cell data such as soil types, crop categories, etc. The other resampling methods will create nonexistent data types if used with raster cells containing discrete data types. For example, given an input raster depicting soil types 10 and 20, Nearest Neighbor resampling will yield an output raster with cells comprising only one of these two data values. An interpolation-based resampling method, on the other hand, will yield an output raster with cells values of 10, 20, and other intermediate values, such as 15. Even if soil type 15 was found in this study area, this soil type does not reflect a transition between soil types 10 and 20 under any circumstances.

3451.7.2

Bilinear Interpolation Method

The Bilinear Interpolation method estimates new cell's value within a 2 x 2 neighborhood of cells. It is based on the presumption that the value being sought lies on or near a straight line joining two known values. This method may be used in resampling a raster object with a different cell size and /or orientation.

Bilinear interpolation is suitable for use with continuous data types where greater brightness represents cells with more of some quantity, such as elevation or temperature. It is not suitable for use with discrete data types such as soil or crop type. It also should not be applied to 8-bit color rasters, classification rasters, or any other raster objects containing similar categorical data.

Bilinear Interpolation often produces results which are acceptably similar to that produced using Cubic Convolution; however, the Cubic Convolution method generally produces sharper results. However, the Cubic Convolution method is more computationally complicated than Bilinear Interpolation, and therefore, the process takes longer.

3451.7.3

Cubic Convolution Method

The Cubic Convolution or Cubic Interpolation method is a computationally intense interpolation method used in raster resampling. Rather than assuming a linear relationship between two known cell values, this method determines a new cell value by fitting a cubic polynomial surface to a 4 x 4 array of cells, generally producing a smoother output raster.

Cubic Convolution is suitable for use with continuous data types where greater brightness represents cells with more of some quantity, such as elevation or temperature. It is generally not suitable for use with discrete data types such as soil or crop type. It also should not be applied to 8-bit color rasters, classification rasters, or any other raster object containing similar categorical data.

Bilinear Interpolation often produces results which are acceptably similar to that produced using Cubic Convolution; however, the Cubic Convolution method generally produces sharper results. However, the Cubic Convolution method is more computationally complicated than Bilinear Interpolation, and therefore, the process takes longer.


Previous Page    Next Page    Top of Page    Table Of Contents    Index    Search