Principal Component Analysis

Principal component analysis(PCA) is very important and useful in performing classification on multispectral or hyperspectral images and displaying high quality three-band overlay color composite images. Because satellite data usually have and tend to have more than three bands, displaying a 3-band composite image only use the information from the selected three channels. Therefore, the composite image cannot represent all the information from all the channels, and is dependent on the knowledge of the human analyst. However, principal component analysis will concentrate most of the information from all the channels to several handable channels, largely improving the 3-band overlage image quality and reducing computational time for image classification and other purposes. The following images display the three-band overlay composite image before PCA and after PCA, respectively. The used image is from Landsat TM with seven channels.

(a) Three-band overlay composite image (b4/red, b3/green, b6/blue) before principal component analysis.	(b) After principal component analysis, the first 3 bands contains 77%, 20%, 1.3% of the total variance in the image. The above figures shows the three-band overlay composite image (b1/red, b2/green, b3/blue) after principal component analysis.