Generate synthetic images with controlled variation using the few sample images.

Hello,

I have an image of a copper wire with defect. I am wondering if it possible to generate synthetic images using this sample image to help me serve as a data set for deep learning applications. Attached is the image for reference. As far as image variations are concerned I am flexible as long as it looks like something that I have attached - Basically a wire containing a defect.

Will be glad if any help can be offered. Thanks a lot.

Answer 1

[cr]

If by defect you mean the circular feature on the left of the image, you can check its attributes (pixel colours) and copy them to random places in the image. However I'm doubtful if you would get a useful training set that way. Genereally when data set is sparse it's recommended to uuse traditional image processing techniques like blob-analysis, at least as a stop-gap measure until the data set is rich enough for training a neural net.

Regards.

[Anand Ra]

Hey, Thanks for the response. Yes, I am referrign to the pin hole feature on the left of the image. I am super new to image processing/ ML, so wondering how to get strarted with your suggestion of copying the attribute and pasting them randomly. Thanks!

[cr]

If this is a one time task it might be faster on a tool like Photoshop. In Matlab you can create masks using from a selected region. Check out drawcircle() and similar draw~ functions. In general, see ROI-based processing.