Skip to main content
. Author manuscript; available in PMC: 2010 Sep 23.
Published in final edited form as: Neural Comput. 2003 Feb;15(2):349–396. doi: 10.1162/089976603762552951

Figure 3.

Figure 3

Comparing the coding efficiency of complete and 2× overcomplete representations on 8×8 pixel patches drawn from natural images. The points on the curve are the results from different values of p, at the bottom right, p = 1.0, and at the top left, p = 0.5. For smaller p, the overcomplete case is more efficient at the same level of reconstruction error (RMSE).