Image Exploration
Contrast of an image is important in distinguishing its visual features. The following picture is of the IDRISI software and the image is from a satellite being used for study. As can be seen, it is displayed in the gray scale palette yet most of the image is the same medium gray, leading to poor contrast. Fine details are hard to pick out, like roadways or other urban infrastructure. The histogram indicates that most of the pixels fall in the 90’s, leading to this overall gray color and that the actual maximum is 190, not 255. For more contrast and better viewing capabilities, the display needs to be stretched so that all colors of the palette, from black to white (255), are used.
One way to change the contrast is to click the auto scale button in the layer properties box. A difference of color (or tone) is noticed but not a change in contrast. The data values themselves are also not changed.
One way to change the contrast that also manipulates the data is to use the STRECH module. Saturation points are created by setting new minimum and maximum thresholds so that all values below or above those numbers become either a black or white color. The remaining pixels will then be stretched across the rest of the palette colors. From the picture below, we can see more contrast in the image to the right of urban areas (dark grey), water (dark), and forested areas (light grey to white).
Supervised Classification
With supervised classification, the user develops the
spectral signatures of known categories, such as urban and forest, and then
the software assigns each pixel in the image to the cover type to which its
signature is most similar. The steps for supervised classification may be summarized as
follows: 1. Locate representative examples of each cover type that
can be identified in the image (called
2. Digitize polygons around each training site, assigning a unique identifier to each cover type;
3. Analyze the pixels within the training sites and create spectral signatures for each of the cover types; and
4. Classify the entire image by considering each pixel, one by one, comparing its particular signature with each of the known signatures. So-called hard classifications result from assigning each pixel to the cover type that has the most similar signature. Soft classifications, on the other hand, evaluate the degree of membership of the pixel in all classes under consideration, including unknown and unspecified classes. Decisions about how similar signatures are to each other are made through statistical analyses. There are several different statistical techniques that may be used. These are often called
classifiers.Completed steps 1 and 2 can be seen below. I created polygons around areas identified to be of a certain type. The digitizer button, which looks like this was used. The polygon features were then added to the h87th4 vector layer.
Step 3 included creating signature files containing statistical information about the reflectance values of the pixels within the training sites for each class. To do this, the module MAKESIG was used. The top graph shows the average pixel number of each training site for the 7 bands used. Bands 4 and 5 are the clear winners because they differentiate the land-uses best.
Principal Components Analysis
One application of PCA is data compaction, This allows you
to keep the most important information while discarding a large proportion
of the data. This was very useful when processor speeds and disk capacity
were low, so it really isn't an issue now. The following are values generated by PCA module in IDRISI.
The Correlation Matrix values indicate how well each band correlates to each
other, values near 1 is good correlation. The eigenvalues
express the amount of variance explained by each component and the
eigenvectors are the transformation equations. The loadings refer to the
degree of correlation between these new components (from the PCA module) and
the original bands.
As can be seen from the data, component 7 does not relate much to any of the bands. The image below shows what component 7 looks like. What can be seen is 'noise', there is little detail or purpose for it so it may be discarded. Component 1 looks much like the infrared image while component 2 is similar to the red image (though it is displayed in black and white). Clearly these two bands contain much important information that would be useful for further analysis.
Unsupervised Classification
Unsupervised classification is another technique for image classification. In this approach, the dominant spectral response patterns that occur within an image are extracted, and the desired information classes are identified by means of ground truthing. To do this the CLUSTER module is used.
This is the original image and then it is classified into broad and fine clusters (following image) clusters. The broad and fine generalization levels use different decision rules when evaluating the frequency histogram for peaks. In broad clustering, a peak must contain a frequency higher than all of its nondiagonal neighbors. Fine classification allows a peak to have one non-diagonal neighbor with a higher frequency. This accommodates true peaks which are otherwise missed because nearby peaks of greater magnitude obscure the usual dip between the peaks
This final image is of the fine clusters classified into their land cover categories. This was done by comparing them to other images created during the supervised classification exercise.
