how can choose threshold method for image with four intensity level?

1 view (last 30 days)
hi all I want to choose threshold method for image with four intensity level. this is my histogram of my image:
which threshold method is better for my image? I mean if my image has The low number of intensity level which thresholding method is better? and why?

Answers (2)

Image Analyst
Image Analyst on 28 Apr 2016
Try this
binaryImage1 = yourImage == value1;
binaryImage2 = yourImage == value2;
binaryImage3 = yourImage == value3;
binaryImage4 = yourImage == value4;
You can tell what the values are from the histogram or using unique(yourImage).

Sign in to comment.

Walter Roberson
Walter Roberson on 28 Apr 2016
Nearly any JPEG image is going to have more than 4 levels. JPEG uses lossy compression that is not able to exactly reconstruct images You will get aliasing effects at edges, pixels that are not exactly what you want.
You should use multithresh and imquantize . The result will be a labeled image. You might want to subtract 1 from the labels so that the black becomes 0 so that it can be ignored afterwards.
Image Analyst
Image Analyst on 29 Apr 2016
multithresh was only needed because you jpegged the image and basically ruined it. It is not needed to segment out the brain. Look at the histogram in my screenshot. See how it's not 4 spikes like in your original histogram picture you posted? The spikes have become widened because the jpegging introduced several new gray levels. We not only have the 4 original ones, but we have a bunch more. Your image got changed. This is why you never use jpeg format for image analysis if you can at all help it. multithresh basically figures out the thresholds to produce an image with the 4 original gray levels. If you had saved as a PNG file, you would not have had that problem.

Sign in to comment.


Find more on Modify Image Colors in Help Center and File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!