How do I apply multi-OTSU to learn from a dataset of retina images the optimal threshold to segment a particular feature? This is important because of variations in images, so that the optimal threshold can be generalized to segment unseen images.
Thank you very much for this tutorial. I can see that you have done very good lessons. I have a question regarding threshold. I have a set of images and I used Tesseract to extract text. But different images give good results depending on the threshold level. So I had to manually change the threshold and check the OCR output, Is there a way that I can get this threshold value automatically for each image?
sir i has problem with module multiotsu error message : cannot import name 'threshold_multiotsu' from 'skimage.filters' (C:\ProgramData\Anaconda3\lib\site-packages\skimage\filters\__init__.py)
Hi, thank you for the tutorial. I'm having trouble running it with a number of thresholds bigger than 6. it takes forever to run. Is there another algorithm that can do this in a shorter time? I know in Matlab its work fast! thank you
I haven't tried with that many thresholds. Did you try k-means or GMM? If they do not yield good results then you may have to look into machine learning. I recorded videos on all these topics, just look at my channel. Hopefully you'll find something that works for you.
Image format doesn't matter as once they are imported to python, they are just a bunch of numbers. For .tiff, you can use tifffile library to load your images. skimage also allows for tiff file import.
Dear Professor, one question. I humbly noticed that these outputs are missing the colors of the legend. How would you suggest plotting the properties of an image file in python, and hypothetically even the % of each color over the entire frame. Coding sources around stackflow don't help effectively. Thank you so much for this possibility!
Great tutorial. I can't import threshold_multiotsu though, I'm running skimage version (0.17.2). Do you know what the issue might be? This would really make my life easier for a school assignment
It is supposed to work for skimage 0.17.2. Please check the spelling, may be you made a mistake. Here is their documentation for that version. scikit-image.org/docs/stable/auto_examples/segmentation/plot_multiotsu.html
Please check if your image being properly imported as a numpy array (plot it after reading), check if the histogram looks ok, check if otsu threshold values actually make sense for your image, finally check what the range of values are for the thresholded image. Also, is it black when you save it as an image and open with Windows or other viewer? If so check if you are saving image with 8 bit values or floating. Many things to check and I hope you can go through them methodically. These steps are important for any image processing, not just otsu.
thanks for this tutorial! Short comment re installation for beginners like me: I had old scikit-image in my Spyder IDE, and 'pip install...' didn't work. To update scikit-image from Anaconda use 'conda install -c conda-forge scikit-image' as suggested here scikit-image.org/docs/stable/install.html Also, multiotsu didn't work for 16bit image, so convert to float
Great video and congratulations on 2k subscribers!
Awesome videos, congratulations on 2k subscribers :)
How do I apply multi-OTSU to learn from a dataset of retina images the optimal threshold to segment a particular feature? This is important because of variations in images, so that the optimal threshold can be generalized to segment unseen images.
Thank you very much for this tutorial. I can see that you have done very good lessons. I have a question regarding threshold. I have a set of images and I used Tesseract to extract text. But different images give good results depending on the threshold level. So I had to manually change the threshold and check the OCR output, Is there a way that I can get this threshold value automatically for each image?
Astounding job sir please make tutorial on brain tumor segmentation using GAN.
thank you very much my question is what is the different between multi otsu and k means and which is the best
I tried to run the code shown in 6:53, but I got a black screen for my multi-Otsu result. What did I do wrong?
sir i has problem with module multiotsu
error message :
cannot import name 'threshold_multiotsu' from 'skimage.filters' (C:\ProgramData\Anaconda3\lib\site-packages\skimage\filters\__init__.py)
multiotsu is not available on older versions on scikit-image. You need at least 0.16.1.
I ve 15 version only... How to update it
Hi, thank you for the tutorial.
I'm having trouble running it with a number of thresholds bigger than 6. it takes forever to run.
Is there another algorithm that can do this in a shorter time? I know in Matlab its work fast!
thank you
I haven't tried with that many thresholds. Did you try k-means or GMM? If they do not yield good results then you may have to look into machine learning. I recorded videos on all these topics, just look at my channel. Hopefully you'll find something that works for you.
What all image formats can be used as input for this code? I want to do otsu thresholding on ".tiff" image. Can the same code be used?
Image format doesn't matter as once they are imported to python, they are just a bunch of numbers. For .tiff, you can use tifffile library to load your images. skimage also allows for tiff file import.
The Otsu method is the same that the multiple otsu with only 2 classes a one threshold?
If number of classes = 2 then multi-otsu should be same as otsu.
Dear Professor, one question. I humbly noticed that these outputs are missing the colors of the legend. How would you suggest plotting the properties of an image file in python, and hypothetically even the % of each color over the entire frame. Coding sources around stackflow don't help effectively.
Thank you so much for this possibility!
Same question i have
Great tutorial. I can't import threshold_multiotsu though, I'm running skimage version (0.17.2). Do you know what the issue might be? This would really make my life easier for a school assignment
It is supposed to work for skimage 0.17.2. Please check the spelling, may be you made a mistake. Here is their documentation for that version.
scikit-image.org/docs/stable/auto_examples/segmentation/plot_multiotsu.html
@@DigitalSreeni Not sure what the problem is, no issues importing anything else from skimage.
really good tutorial, thank you :)
You are welcome :)
Sir pls i have a question, is it possible to do transfer learning for u-net ?
Yes. I did not cover it in any of my videos but you can find many resources online.
you are great!!
very useful , thanks !
Hi!, My multiotsu segmented image got black... any idea?
Please check if your image being properly imported as a numpy array (plot it after reading), check if the histogram looks ok, check if otsu threshold values actually make sense for your image, finally check what the range of values are for the thresholded image. Also, is it black when you save it as an image and open with Windows or other viewer? If so check if you are saving image with 8 bit values or floating. Many things to check and I hope you can go through them methodically. These steps are important for any image processing, not just otsu.
I think that is impossible by OpenCV, right?? I spent more time but fail.
I'm not aware of opencv having multiotsu functionality.
thanks for this tutorial!
Short comment re installation for beginners like me:
I had old scikit-image in my Spyder IDE, and 'pip install...' didn't work.
To update scikit-image from Anaconda use
'conda install -c conda-forge scikit-image'
as suggested here scikit-image.org/docs/stable/install.html
Also, multiotsu didn't work for 16bit image, so convert to float
Thanks for the info, hopefully it helps newbies.