How to increase the number of samples for training?

I am using SVM for skin disease classification but have fewer number of images and each image has more than one lesion so how I can take every lesion as a separate sample in order to increase the number of samples for training.

Any help is greatly appreciated.

Answer 1

[Dinesh Yadav]

I think you can simply crop and save one lesion at a time. This will give you increased no of samples. Also you can use data augmentation. Lets say you have original image with 10 lesions you can crop it in any way having 10 images with 1 lesion each or 2 images with 5 lesions or any such combination and use data augmentation on cropped images also to increase no of training samples.

[Mohammad Farhad Aryan]

Thank you Dinesh Yadav for your answer.

Actually, some papers used block processing and some other techniqes to consider each lesion as a separate sample in order to increase the number of samples then extracted texture features of each lesion using GLCM. I want to implement such kind of techniques to increase the number of samples and extract texture features of each lesion using GLCM but don't know how to implement it.

Help me please!

[Dinesh Yadav]

https://in.mathworks.com/help/images/block-processing-large-images.html

https://in.mathworks.com/help/images/ref/blockproc.html

https://in.mathworks.com/help/images/distinct-block-processing.html

https://in.mathworks.com/help/images/parallel-block-processing-on-large-image-files.html

These are the articles on block processing in MATLAB. This divides images into multiple segments based on the size you give. Is this what you meant?

https://in.mathworks.com/matlabcentral/fileexchange/22187-glcm-texture-features

For feature extraction using GLCM.

Hope it helps.

[Mohammad Farhad Aryan]

Thanks Dinesh Yadav for your help!

I am going forward with blob detection using regionprops which I think some papers have used it.