1
$\begingroup$

I have been experimenting with computer vision techniques to find defects in paintings, more specifically defective brush strokes. I have tried a handful of techniques such as different filters, One class ML classifiers. image subtraction etc. The problem seems to be the orange-peel kind of textural noise in the image (present even in perfectly painted ones). Any idea how to go about finding the defected brush strokes and highlighting them? Pictures attached for reference

Defective wall paintPerfectly painted wall

$\endgroup$
  • 1
    $\begingroup$ I'd argue the rows of the lower picture aren't as highly correlated as the rows of the upper picture. I assume you've looked at basic techniques like autocorrelation, inter-row correlation before firing the ML cannon at the problem? $\endgroup$ – Marcus Müller Jan 16 at 16:27
  • $\begingroup$ I haven't tried autocorrelation or inter-row correlation, but would it work every time as the brush strokes might be different each time? $\endgroup$ – Bikram B. Jan 17 at 8:04
  • $\begingroup$ yes, because you're not comparing to a "prototype" brush, but to the row above. $\endgroup$ – Marcus Müller Jan 17 at 9:23
  • $\begingroup$ I'll give that a try and let you know if it worked! $\endgroup$ – Bikram B. Jan 17 at 9:40
  • 1
    $\begingroup$ The first image is definitely much more periodic than the second. I'd also suggest that this periodicity would show up very intenesely in the frequency domain. $\endgroup$ – A_A Jan 20 at 11:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.