Abstract

In this paper, we generalize the idea from the method called "PCANet" toachieve a new baseline deep learning model for image classification. Instead ofusing principal component vectors as the filter vector in "PCANet", we usebasis vectors in discrete Fourier analysis and wavelets analysis as our filtervectors. Both of them achieve comparable performance to "PCANet" in benchmarkdatasets. It is noticeable that our algorithms do not require any optimizationtechniques to get those basis.

 

Quick Read (beta)