I have a neural network which can classify gray images and now I want to use it to classify color ones. How can I do it?
In my previous network if the output was more than 0.5 the image was a member of class else not. How about color image? Here I have an output vector (RGB or HSV or ...) instead of scalar number.
Hello Md Amiri, Please go through these steps which helps you to work with color images using neural networks for classification in matlab, so i think this will help you for your work, ok thank you.
function scan(img)
files = dir('*.jpg');
hist = [];
for n = 1 : length(files)
filename = files(n).name;
file = imread(filename);
hist = [hist, imhist(rgb2gray(imresize(file,[ 50 50])))]; %#ok
end
som = selforgmap([10 10]);
som = train(som, hist);
t = som(hist); %extract class data
net = lvqnet(10);
net = train(net, hist, t);
like(img, hist, files, net)
end
function like(im, hist, files , net)
hs = imhist(rgb2gray(imresize(im,[50 50])));
cls = vec2ind(net(hs));
[~, n] = size(hist);
for i = 1 : n
if(cls == vec2ind(net(hist(:, i))))
figure('name', files(i).name);
imshow(imresize(imread(files(i).name), [100 100]))
end
end
end
in color image there are several kind of information you may use:
1- the color information, using RGB vlaues, but remember that they highly affected by the illumination, so one can use other color systems such as HSV, or use R/(R+B+G) for red components and same for B and G.
2- texture information: normally texture is calculated on grey scale images to get the moments, entropy and other statistics, try to use such statistics on color information.
3- transformed information, Fourier, DCT, wavelet ... those also designed for grey scale images, but they can be used on color images.
Get your features, train and test ANN using standard databases, compare your results with other methods and you will be there.
Dear Mohammad,
I like the Joachim' idea about the combination of different spaces (RGB or HSV). In this regard, I suggest you to add a fuzzy layer to combine the spaces based on your own definition. depending on your experience and sense about the dataset you can define suitable membership functions and reach better results.
First, you will have to clarify the objective of classification in case of color image. As far as your comment on your existing gray classifier, it seems to be a 2-class problem, i.e. object or no object. In case of color image classification also, you will have to specify the no. of classes you are trying to identify. Then according to the object definition, you can develop feature set for the classifier. In this feature set, you can use color component values. Normally, HSV model is used for true color identification. But you can use RGB model also depending on the problem domain. Then you can train the ANN and then test it on real data.
Hello
I would suggest to repeat the process you've done with Grey images to each band based on the colour space you are using(RGV, HSV, XYZ ..etc.) and then you can combine the results rather than converting the colour image into grey image. I am not sure what features you are using but histogram statistical measures may work well in your application.
all the best....
There are several approaches where pixel values are used to classify it as a part of an object
- Badges
- Science topic