Multilevel image thresholding combining distinct colors

I am trying to segment an image. It has 8 distinct colors (regions) excluding the background white. I am using multilevel thresholding with 8 levels out of which 6 regions are segmented properly. However, the method combines two regions which is highlighted in circles. Clearly, the grayscale image (middle) also shows the difference for this region. I am not sure how to get this regions segmented along with the rest of the regions. Any help will be much appreciated. I am also reproducing the code below.

% Read the image and convert into grayscale
I = imread(filepath)
I = rgb2gray(I);
% Define thresholds for the 8 distinct regions excluding the white background
thresh = multithresh(I,8)
% Segment the image and display
seg_I = imquantize(I,thresh);
imshow(seg_I == 6)

[KALYAN ACHARJYA]

There are distinct difference between those two labels in the circles (2nd Image).

[Kasthuri Kannan]

Yes, there is - actually I managed to find out the actual gray level for those regions and used the values to segment them, instead of the method to automatically determine them.

Answer 1

[Kritika Bansal]

Hi,

I tried replicating the issue on my end, but the output comes out to be correct in my case when I ran the following commands on MATLAB Online R2019b.

These are the commands I used:

clear all; clc; 
I=imread('input.png'); 
imshow(I); 
I = rgb2gray(I); 
% Define thresholds for the 8 distinct regions excluding the white background 
[thresh, metric] = multithresh(I,8) 
% Segment the image and display 
seg_I = imquantize(I,thresh); 
figure, imshow(seg_I, []); 
figure, imshow(label2rgb(seg_I)); 

Following is the output:

thresh = 
1×8 uint8 row vector 
23 61 90 116 146 168 182 219 
 
metric = 
0.9982 

I am also attaching the input and output images for your reference.

[Kasthuri Kannan]

Thanks. I think the R2019b solves this -

[KALYAN ACHARJYA]

+1

Truly