How to reduce brightness of ceratin pixels from an Image?

Hi,

I have a set of particle micro-CT images for 3D structure generation. However, the ash on the particle cause pixels to be super brightness and brings negative effect on threshold value selection for segmentation. The basic idea is to set a value of brightness to be a threshold value, any higher brightness value of each pixel will be reduced to a lower brightness value. But how to appropriate determine those brightness values is a problem. Is there any methods or functions in Matlab can solve this problem?

Thanks!

Answer 1

[Bjorn Gustavsson]

Is this the operation that you're asking for?

too_bright = 123; % or your chosen level
replacement_level = 123; % Not a good idea to give it a lower value, you might want a nan to flag pixels as bad?
Im(Im(:)>too_bright) = replacement_level;

Where Im is your image.

HTH

[Dongyu Liang]

Hi,

Thank you for the reply.

Yes, it is the basic idea. But I dont know if the chosen level is appropriate, and it will take a lot of time to test manually. So I am wondering is there a well-developed method or function to reduce brightness without strongly affecting the threshold value selection for segmentation?

Thanks!

[Bjorn Gustavsson]

You get have some understanding of what the outliers might be by looking at the intensity-histogram:

hist(Im,min(Im(:)):max(Im))

That way you'll see the image in a different way. If the hot pixels are true outliers then you should be able to separate them from that plot. It seems difficult to give more specific information about this without knowing the imaging system, the imaged objects etc. You know these things, you'll be the one to make these decissions, so my best advice is to look at the images. You can also just plot the image:

plot(Im)

or simply look at the images:

imagesc(Im)

...and then modify the color-axis:

caxis([min(Im(:)) 321])

interactively and see what that does to the images. Just look at the data, in a multitude of ways to learn what's going on.

HTH

[Dongyu Liang]

Hi,

I have found this example

https://dsp.stackexchange.com/questions/1215/how-to-remove-a-glare-clipped-brightness-from-an-image

which seems what I want. Based on the comments, there was no Inpaint funcution in Matlab at that time. I am wondering if there is similar function available now in Matlab?

Thanks!

[Bjorn Gustavsson]

What are your original images? Are they slices of a 3D-tomography of sorts? Or are they the images you or someone else will use for tomographic reconstruction?

If it is the latter definitely never ever use inpainting or any other "magicing up" of image information. Typically when solving inverse problems like CT we have problems with error and noise-propagation already with "perfect" observations. To come up with any ever so clever fake observations where your input data is corrupted is not the way to go, just follow Wittgesteins sound advice "Wovon man nicht sprechen kann, darüber muß man schweigen". If you have no good data from a pixel remove it from the processing, don't lie about what should've been there.

If you have slices that you're going to use for some sort of quantitative data analysis of the CT-ed object then do that with the good information, and still avoid to invent information that is not there, you be far more true to you observations if you learn to handle the outlier-problem.

If you have slices of some 3D-CT and want to display some pretty visualizations for show - then knock yourself out no-one is going to be bothered about it anyway, but know what you've done and what is what.

I don't know what Inpaint in matematica does, but "texture synthesis" is mentioned. I've seen that mentioned but Ḯve never bothered with that. There are a Poisson-type inpainting function on the file exchange, that I use: inpaint_nans.

HTH

[Dongyu Liang]

Hi,

Thank you so much for the advises.

The ash part is not what I want to remove or lie about. In my project, I want to deal the ash as a part of particle but it gives super brightness pixels which make the threshold method inaccurate for segmenting. For the particle without ash, I can get binary image for entire particle with good detail inside. But for the particle with ash, only ash part can be displayed in the binary image whatever threshold methods I use. That is why I want to reduce the level of ash part close to the level of particle part.

Thanks!

[Bjorn Gustavsson]

Well, if the ash-part is just a small super-bright part of the particles you want to identify, then by all means simply replace these outliers with nan and then use the inpaint_nans function to fill in those wholes. That should give you a non-toxic (for your processing) filling-in of your particles. Then the segmentation/identification should(ought to?) work OK. Just take extra care to check that nothing "peculiar" happens with the inpaint-threshold-identification sequence - I always think about this in terms of making a decision about radioteraphy on a brain-tumor, if I were to have done some inpainting type operation I'd like to be pretty darn sure about the advice I'd give (in my field the stakes are not that serious, but I assume that yours is...).

[Dongyu Liang]

Thank you!