The JavaMet applet provides a visualization tool for student Meteorologists learning the concepts of classification for N-dimensional satellite images. In the past students have been taught classification using 2D examples, and have struggled to extrapolate the concepts to 3, or more, dimensions. It is hoped that by providing a 3D-visualisation and classification tool the students will gain a deeper insight into image classification for higher dimensions.

The illustration shows a sample classification space. On the left is a three-channel satellite image displayed in RGB format. On the right is the classification space. The three axes correspond to the red, green, and blue components of a given pixel. Each Voxel (3D pixel) within the cube is drawn using the appropriate color for its RG and B components.

In addition to the RGB values present within the image, information about their frequency can also be calculated. Certain RGB values may be more numerous, in terms of pixel-count, than others. The frequency information allows a threshold to be applied to the color-cube, and only the Voxels falling within the threshold limits are drawn. This will allow the student to assess how much of a color is present in the satellite image.

In the illustration above only the high frequency RGB values have been drawn and two distinct clusters have emerged. To the lower left are the brown shades that correspond to the area of northern Africa, while the blue shades at the upper right of the cube correspond to the atmosphere around the earth. In this way the colour-cube can be used to visualize and classify RGB ranges as corresponding to particular features in the image.

• Daniel Selman
• Willem Moolenburgh
• Todd Densmore

• Adrian Fletcher