Path: news.mathworks.com!not-for-mail
From: "Jeevashini" <vashini13@mathwork.com>
Newsgroups: comp.soft-sys.matlab
Subject: Neural Network Training for Image Classification
Date: Tue, 27 Jan 2009 14:14:01 +0000 (UTC)
Organization: Informatics Institute of Technology
Lines: 19
Message-ID: <gln4r9$cqh$1@fred.mathworks.com>
Reply-To: "Jeevashini" <vashini13@mathwork.com>
NNTP-Posting-Host: webapp-03-blr.mathworks.com
Content-Type: text/plain; charset="ISO-8859-1"
Content-Transfer-Encoding: 8bit
X-Trace: fred.mathworks.com 1233065641 13137 172.30.248.38 (27 Jan 2009 14:14:01 GMT)
X-Complaints-To: news@mathworks.com
NNTP-Posting-Date: Tue, 27 Jan 2009 14:14:01 +0000 (UTC)
X-Newsreader: MATLAB Central Newsreader 1171339
Xref: news.mathworks.com comp.soft-sys.matlab:514232



Hi,

I am working on image classification based on colour histograms. I have 4 classes of images. Each class of images have a different colour dominant in them. For each image I have HSV histogram. The HSV histogram is made from 8 bins for Hue, 2 bins each for Saturation and Value. Therefore each image has a vector of 32 data.

I have 65 training images in each of the 4 classes: total of 260 images.
I need to feed the 32x260 matrix to train the neural network.

My objective is to compare two methods in classification. One is classifying the image as a whole. Other is dividing an image into blocks and classifying them individually, then finally based on some voting method or some other method classify the image as a whole.

I have basically two questions in carrying out this process.

1. Does having 8x2x2=32 HSV histograms suffice for the training process?
2. For the second method of classification, do I have to perform training separately on the individual blocks? Or can I use the trained network in the first method(whole image classification) for classifying the blocks?

Some advice would be helpful, please.

Thanks,
Vashini