USGS MIPS PCA Documentation

USGS MIPS PCA Doumentation

This program transforms a set of input multi-band images into its principal components. Typically, a set of multi-band images will have a high degree of correlation, thus having redundant information within the data set. Transforming the set of images into its principal components results in a data set where the output images are uncorrelated, thus eliminating the redundancies.

The first principal component image contains much of the image variance, with successive principal components containing less and less variance. Generally, if several images are used as input the last principal component image will contain mostly noise and little useful information.

Files

This program requires as input an eigenvector file and two to seven unsigned 8-bit image files. It will generate the same number of output image files as used as input, and the data type will be the same(i.e., unsigned 8-bit image files). The following information is transferred from the input files to the output files(if appropriate):

Image labels
Mosaic labels
Projection labels
Sonar labels
Navigation file
History file

The eigenvector file must be produced ahead of time using the program STATS. Also, the input image files must be entered in the same order for both the STATS and PCA programs in order to obtain proper results.

Parameters

MIN and MAX: These parameters are arrays used to define the valid dn range to be used from the input images. Element one of each array corresponds with input image one, element two with input image two, etc. Thus each input image has its own min/max pair. Prior to applying the principal component transformation to each image pixel , the pixel's dn values are examined to see if it they are in the desired/selected min/max range. If the dn values of all the input images are valid the transformation is applied, otherwise all DNs at that pixel location in the output principal component images are set to 0. The default values are 0 and 255 for all input images.
OFFSET: This parameter is an array of offset values that are used to keep the results of the principal component transformation within the 8-bit range of 0 to 255. The default values are determined by the program STATS that is required to generate the eigenvector file. The first element is the offset for the first principal component(generally negative), and the second, third, etc, elements are the offset values for their respective principal component images, with default value of 127 used for these component images.

Uses

Principal component analysis has several uses within remote sensing. First, it can be used prior to classification for data set volume reduction. For example, using six images as input to PCA, the first three principal component images will contain the large majority of the total variance, so they can be used as input to classification procedures with little loss of information.

Another use of PCA is what we call selective principal component analysis. It uses various multispectral band combinations as input to principal component analysis in order to generate a three image combination that minimizes further the loss of information and can be used to generate a color composite with maximum information content. For example, using bands 1, 2, and 3 from a Landsat TM image as input to PCA; then TM bands 5 and 7 as input to a seperate PCA; a maximized information three image set would be the first principal component of TM bands 1-2-3, TM band 4, and the first principal component of TM bands 5-7. This could also be used to generate a color composite that would contain this maximized information image product.

Examples

From left to right, original Landsat MSS Bands 4/5/6.

[Example Image]

From left to right, first, second, and third principal components.

[Example Image]

USGS MIPS PCA Doumentation

Files

Parameters

Uses

Examples

See Also