An image is a matrix (actually three matrices, or RGB color values). I read in
an image, 352x192, and computed the SVD of the three color bands. I then reconstructed the image in three ways:
Using all the information;
Using 90/192 (~47%) of the information; and
Using 30/192 (~16%) of the information.
Here are the results:
Original image, using all 192 outer products from the SVD
Compressed using 90 outer products -- it's almost the original -- can you find any differences?
Compressed using only 30 outer products -- quite "lossy"