970304.4.vis1.jpg
970304.4.vis2.jpg
970304.4.m50.123.jpg
970304.4.m50.0.123.jpg
970304.4.m50.234.jpg
970304.4.m50.0.234.jpg
970304.4.m50.345.jpg
970304.4.m50.0.345.jpg
970304.4.m50.456.jpg
970304.4.m50.0.456.jpg
This page is under construction. Currently, only some sample images that were generated using these tools are available.
The images below are derived from an AVIRIS data set (flight f970304t01p02_r02, scene 4). AVIRIS data sets are particularly challenging to the analyst because they represent a very large number of observations on a high dimensional space. This instrument records reflectances from 224 narrow-band channels from mid-infrared to blue-green with approximately 10 meter spatial resolution for most flights. In addition, the bands are highly correlated, so de-correlation techniques are a natural first step with this data. However, since this data typically represents a mixture of different features, classical de-correlation methods such as principal components can be distorted by any outlying subgroups. For that reason, a robust singular value decomposition should give a more effective transformation of the image data.
The first row contains false-color images generated from the raw data that approximate what this region looks like visually. The remaining images were generated from a set of 45 bands using the methods described in the IEEE Proceedings paper. The images on the left were developed from various components of the robust singular value decomposition of the image data. The images on the right represent "outlier" maps based on the corresponding components. The second set of images are based on a set of 88 band ratios defined on these 45 bands. Click on a thumbnail image to view the full-size image.
The next set of images were generated using the same tools applied to the data after forming band-ratios.
 970304.4.r.m50.123.jpg |
 970304.4.r.m50.0.123.jpg |
 970304.4.r.m50.234.jpg |
 970304.4.r.m50.0.234.jpg |
 970304.4.r.m50.345.jpg |
 970304.4.r.m50.0.345.jpg |
 970304.4.r.m50.456.jpg |
 970304.4.r.m50.0.456.jpg |