Description

This animation shows PCA in a new light--a form of signal processing. We begin with a gray scale image in the upper panel. Each pixel is a value between 0 and 1 that represents white, black and shades of gray in between. Each pixel can be viewed as representing a value of a correlation. The combination of gray scale pixels emerges as an image of a human face.

The middle panel presents the results of a PCA where the first image is only factor 1, the second image is factors 1 and 2 superimposed, the third image is factors 1-3 superimposed, etc.

The third panel shows each factor individually rather than superimposing over the previous factors. This shows that as the factor number increases the signal information involves higher frequencies. The earlier factors are pretty broad-stroked factors that show general regions of white and black. The higher factors add the high frequency special detail such as the shading around the nose, etc., which gives the blobs emerging from the first few factors the distinguishing features of a face.

Richard Gonzalez, University of Michigan

gonzo at umich dot edu

Click here to give feedback on this site or report problems.