How to make equal number of images in each class.? which is the best method for this.

Say i have different classes of images, but they are not equal in number to train. some class of images are in 1000's, some are in 100's, some are below 50 and 10. How i can make them equal.? Which is the best method for this.After that i need to do feature extraction. So if i have good number of images in each class i can train well i guess.

[Sayyed Ahmad]

could you post some code to better understanding of your problem please.

[KSSV]

What classes are they are of? May be you can google in images and collect more images if they are available online.

[Karthik K]

They are medical images. Not available online. x ray images.

[Karthik K]

@Sayyed Ahmad, i didnt do any coding. i needed some ideas or methods how we can do that. Like either duplicating (replicating) with some little rotations. so that training would capture much more efficient features. how to do i am not getting any idea.

[KSSV]

YOu can't get x-ray image sin google? I think you shall get those.

[Karthik K]

if 10 classes i cant try it out friend. but here it is more than 100 classes. to make all of them equal its a bigger task. so thinking ti make it duplicate in with some rotation methods if any.

Answer 1

[Image Analyst]

Try randsample() or randperm() to get the indexes of images to use.

[Karthik K]

Both does not work.. This is the sample attached image file. Like these i have different parts of the body x ray images. but all are not in equal number. to train it efficiently i want to make all class of images in equal number. so classes where i have less images i need to duplicate it with some rotations. how to do it. i think i am clear now with my question.

[Image Analyst]

Of course it works. If you get directory listings of all the files of the various classes, and you wanted, say, 50 files of each class, you could get their indexes like this

class1 = dir('class1*.png'); % Get listing of all class 1 images.
class2 = dir('class2*.png'); % Get listing of all class 2 images.
numImagesToExtract = 50;
randomClass1Indexes = randperm(length(class1), numImagesToExtract);
randomClass2Indexes = randperm(length(class2), numImagesToExtract);

Now you have the indexes of the class* variables to use. Just use those indexes to get the filenames of the randomly chosen files to use. What did you do differently?

By the way, you did not make your question clear now. Saying "some class of images are in 1000's" is completely vague. This could mean "some classes have 1000's of images", or "in some classes, the feature values we extract have value in the 1000 to 10 thousand range" or any number of other interpretations. I have no idea what you intend, but my code extracts the same number of randomly-chosen image files from variables having file names in the different classes.

[Karthik K]

Sorry if my question is not understanding and clear. Here this code extracts the image index values. say if i have 60 images of class hand, out of these 60 images , 50 image index values are taken. But what i actually wanted is to create duplicate images which i have in each class. Say i am having 60 images in class hand. if i want the images to be 100, from 60 to 100 how can i duplicate (create copies of images) from the existing images which i have.

Since x ray images are taken in different angle, mean while i am thinking how can i make some rotation to the existing images when creating duplicate images.