Image Classification
We have talked about how to make features stand out in greyscale and 3-color images. Image classification seeks to discriminate features on the image based on their spectral character. In the following example, the reflectance of sand is greater than trees in one band, and so one could classify them based on the midpoint
However, there is a good bit of overlap. some of which may be due to illumination. The upper end of the tree distribution is on the sunny side of the hill and the lower end of the sand distribution is on the shaded hillsides. Ratios are used, ideally, to separate these two distributions. Principal components other than the first can also be used. No method will reduce overlap completely.
Of course as more criteria are added, it becomes easier to distinguish between separate elements as shown in this two-ratio image.
The best way to evaluate whether images are discriminated enough to classify is to perform a simple discriminant analysis on the features.