3
$\begingroup$

I'm working in a computer vision project, where the goal is to detect some specific parasites, but now that I have the images, I noticed that they have a watermark that specifies the microscope graduation. I have some ideas of how to remove this noise, like detecting the numbers and replace for the most common background or split the image but if I split the image I'll lose information.

I would like to hear some recommendations and guidelines of experts.

Example image below.

enter image description here

Improve this question
$\endgroup$
1
  • 3
    $\begingroup$ Hello! Your topic is quite misleading, since you do not want to denoise your image, but exclude a certain area from your analysis. Generate a mask and only evaluate your image where the mask is True, and have it be False where the label is, $\endgroup$ – M529 Apr 10 '19 at 19:21
0
$\begingroup$

I am also not an expert, but I would consider this: The 50 µm watermark is burned in the image and it is complete black (0 0 0) in rgb. It has a significant influence to any analysis oyu make on the image regarding statistics like mean, median, edge detection etc.

So it needs to be removed. The easiest method to remove it would be to find out if you microscope has a setting to metadata this information instead of burning it in the image. If that is not available and you only have this image provided, then I personally would go as follows:

  1. Segment the black "Noise" watermark area and
  2. use that selection as a mask to fill with the mean color.
  3. Do your analysis. Be critical of whether it influences the results and refine if necessary. (E.g. if it still interferes with edge detection add some random noise).

Either that, or, if you can afford it - exclude the area of the watermark altogether from the analysis.

But please correct me if anyone knows better.

Improve this answer
$\endgroup$
0
$\begingroup$

My sugestions are based on simple approaches (and since I don't know the resolution of the images or if the watermark is always in the same position):

Option 1. use a median filter with a "big" kernel.

Option 2. extract the (near) black pixels (for example p<5) and attribute them a new value based on the nearby colors. (if you want you can only consider the blob with an area greater than a given threshold)

Improve this answer
$\endgroup$

Your Answer

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

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