feature selection for a 2D image
Message-ID:<77a59c5b-0fd3-4ae1-b077-c2bbcfe835bd@w1g2000prm.googlegroups.com>
Subject:
feature selection for a 2D image
Date:Sun, 4 Jan 2009 12:40:26 +0100
After theoretical studies, I want to understand how really ANN work in real applications. For example in character recognition how do we choose features? 1) Have we to find a method for extracting them from image (I've read about methods for finding the number of "lakes", "bays", "lids" and "strokes")? 2) Or we simple have to put the input as it is (for exampke for a 10x10 pixel, 100 inputs) and afterwords reduce the dimensionality with mathematical methods such as PCA? ---- LC
Message-ID:<4ea802d9-26fc-4112-9505-aa0b37c95b0a@z28g2000prd.googlegroups.com>
Subject:
Re: feature selection for a 2D image
Date:Tue, 3 Feb 2009 04:20:01 +0100
On Jan 4, 7:40=A0pm, Livio Carrieriwrote: > After theoretical studies, I want to understand how really ANN work in > real applications. > For example in character recognition how do we choose features? > 1) Have we to find a method for extracting them from image (I've read > about methods for finding the number of "lakes", "bays", "lids" and > "strokes")? > 2) Or we simple have to put the input as it is (for exampke for a > 10x10 pixel, 100 inputs) and afterwords reduce the dimensionality with > mathematical methods such as PCA? > > ---- > LC First you select the feature and then employ ANN to learn how to use the feature . So both above is usable in practical while the performance may be quite different



RSS News Feed