It is often necessary to locate the exact row and column coordinates of a study area within an image during the early stages of a remote sensing project. Many digital image processing systems today are unable to display a full image at the normal commercial pixel scale (>3000 rows and 3000 columns). Being unable to view the entire image may pose a problem in locating the exact coordinates of a study area. Under such circumstances, image reduction allows the analyst to view a subset of an image at one time on the screen by reducing the original image dataset down to a smaller dataset. This technique is useful for orientation purposes as well as delineating the exact row and column coordinates of an area of interest.
To reduce a digital image to just 1/m
squared of the original data, every mth row and mth column
of the image are systematically selected and displayed. An image containing
5160 rows by 6960 columns could be reduced so that every other row and
every other column (i.e., m = 2) were selected for a single band.
This reduction would create a sampled image containing only 2580 rows by
3480 columns. This reduced dataset would contain only 25% of the pixels
found in the original scene (Jensen, 1996). The logic of a simple 2x integer
reduction is shown in Figure 6-1.1.
 |
Figure 6-1.1
Image Reduction |
Unfortunately, a simple 2x integer reduction is often still too large to view on most screens. In cases where a 2x reduction is not small enough, the data must be sampled more intensely. An image sampled at a 10x reduction, meaning every tenth row and tenth column of the image is sampled, will result in an image consisting of 516 rows and 696 columns. Although a resampled image at this scale contains only 1% of the original data, it is an small enough to view the entire scene within the screen. Because a resampled image has obviously lost many of its original pixels, it does not contain adequate data for image processing and interpretation. Resampled images are more commonly used for orientering within a scene and locating the exact row and column coordinates of a specific study area. These coordinates can then be used to extract a portion of the image for full resolution analysis. Figure 7-1.2 shows a simple 1x, 2x, and 4x reduction of a portion of original Landsat TM data of Charleston, SC.
 |
 |
 |
 |
| Landsat TM data of Charleston, SC | Reduced at 1x | At 2x | At 4x |
Digital image magnification is often referred to as zooming. This technique most commonly employed for two purposes:
 |
Figure 6-1.3
Image Magnification |
In many sophisticated digital image processing systems, the analyst can specify a floating point magnification rate such as 2.75x. This requires that the original remote sensor data be resampled in near real time using one of the image resampling algorithms (e.g., nearest neighbor, bilinear interpolation, or cubic convolution). This technique is often used when detailed spectral reflectance or emittance characteristics of a relatively small geographic area of interest is needed. Being able to zoom in to the raw remote sensing data at precise floating point increments can also be helpful during a supervised classification of an image. Figure 7-1.4 shows Landsat TM data od Charleston, SC magnified 1x, 2x, 3x, 4x, and 8x. Note that as magnification increases in the urban areas of the image, building shadows become more apparent.
| |
 |
 |
| TM data of Charleston, SC | Magnified at 1x | Magnified at 2x |
 |
 |
 |
| Magnified at 3x | Magnified at 4x | Magnified at 8x |
