2
$\begingroup$

I would like to test (predict) whether camera images are dusty/dirty or not.; a classification task indeed.

The question is about finding the most appropriate Machine Learning approach, perhaps any suggestions for possible Deep Learning methods specifically tailored for this particular problem?

Improve this question
$\endgroup$
4
  • 2
    $\begingroup$ This is a binary classification task. Read the literature to find relevant features; e.g., Removing Image Artifacts Due to Dirty Camera Lenses and Thin Occluders. Welcome to the site! $\endgroup$
    – Emre
    May 22, 2018 at 17:19
  • $\begingroup$ There was a Kaggle comp for the same named like IEEE...(can't remember exactly) $\endgroup$
    – Aditya
    May 22, 2018 at 19:03
  • 2
    $\begingroup$ What technique to use for a particular problem is based on certain factors which you have not provided information for. Are the images labeled? How many labeled images do you have access to? $\endgroup$
    – JahKnows
    May 23, 2018 at 3:03
  • $\begingroup$ I have over 10 000 labeled images $\endgroup$
    – welu
    May 23, 2018 at 9:00

1 Answer 1

Reset to default
0
$\begingroup$

What kind of distortion does dirt/dust produce? If it is somewhat regular, for example more or less regularly sized speckles (greyish circles) it might be worthwhile to try to classify the logarithm of the real part of the fourier transform of the picture, instead of the original picture.

Also, knowing the physical properties of the distortion might guide you towards devising simple heuristics (for example based on some frequencies in the fuorier spectrum) that are as good as a ml-based approach.

Improve this answer
$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.