Scratchness detection/scoring on plastic/metal surface

ygc 0
Viewed 46

Scratch on plastic surface may have huge/tiny impact on product given the size, color, depth,I call the impact of those factors:"Scratchness".

Lighter color or smaller size of scratch may have tiny impact, we can give it score 1 for scratchness: Darker color and bigger size may have huge impact on the product, we can give it score 5 for scratchness.
I have some example img here with scratchness score:

I am working on a solution to find a image based solution to detect the impact of scratch with noisy background, I can imagines two different solutions:

  1. mathematics based image processing algorithm based on contrast and other characteristic of these imgs
  2. deep learning with proper labeling, either classification or detection/regression. Please shed some light, I can provide more data in case needed

score 3

score 5 score 1

deep-learning image-processing
1 Answers

I can't speak to how well a mathematics based approach will work but I can say that deep learning works well for this.

I've seen high 97+% per-class accuracy on similarly formulated "how scratched" classification problems by fine-tuning a transformer such as CLIP.

Disclaimer: I work for an ML-as-a-service company so I'm probably biased toward ML solutions.

Dan Ott0
Related