What Is Image Segmentation ? And How To Do Image Segmentation Using Python Coding

 What Is The Objective Of This Topic: The objective of this topic is to apply region growing and k means clustering algorithm to segment an image.

 What Is The Description Of This Topic: Region Growing is also another kind of region-based segmentation method. For region growing, a “seed” point is selected. Whenever this seed is encountered in an image, its surrounding neighboring pixels are checked and a decision is made whether a neighboring pixel should be added to the region or not following a selection criterion. One criterion that can be used is that only that pixel is kept from the neighbors that has the same intensity value as the pixel itself.

Several complete passes of the image are needed in order for region growing to completely work. 

K means Clustering:

K means clustering is the most basic and widely used clustering technique for data grouping, in machine learning and in colored image segmentation. The algorithm for K-means clustering is as following 1. Chose the number (K) of clusters and randomly select the centroids of each cluster. 2. For each data point: 

     A. Calculate the distance from the data point to each cluster. 

     B. Assign the data point to the closest cluster. 

3. Re compute the centroid of each cluster. 

4. Repeat steps 2 and 3 until there is no further change in the assignment of data points (or in the centroids). 

After theoretical part we will understand this topic through some practical work using python language.

Here is the practical problem let's try:

1.Write a program that take dermoscopic image as input (images used in assignment 1) and applies region growing on it. You have to segment out lesions. Take center of the image as seed point. Display the resultant image.

Code

import numpy as np

import cv2 as cv

import math as m

img = cv.imread('freestudystrom.bmp',0) # Read input image

dims = np.shape(img)         #image dimensions

rows = dims[0]               #rows of image       

cols = dims[1]               #columns of image



m = np.min(img)              #minimum pixel value          

seed_val = m+6              #seed value   

Epsilon =10                 #epsilon                       



padded_img =np.pad(img,1,'constant')

output= np.copy(padded_img)



def Again(seed,i,j):

    for y in range(i - 1, i + 2):

        for z in range(j - 1, j + 2):

            if (output[y, z] != 255):

                print('Call again',y,z)

                Neigh(seed, y, z)



def Neigh(seed, i, j):

    while (i-1>0 and j-1>0):

        for k in range(i - 1, i + 2):

            for f in range(j - 1, j + 2):

#calculate distance of the neighbouring pixels from the centre one

                diff = abs(int(padded_img[k, f] - padded_img[i,j] )) 

                if diff <= Epsilon:

                    output[k, f] = 255

                else:

                    output[k, f] = padded_img[k, f]



        if (i>1 and j>1):

            Again(seed_val, i, j)

        else:

            break

#main program

for i in range(1, rows):

    for j in range(1, cols):

        if padded_img[i,j] == seed_val:   #centre of region to be detected

            cx =i

            cy=j

print('Seed value at [',cx,'][', cy,']')

Neigh (seed_val,cx,cy)                     #call the function

cv.imshow('Region Growing Algorithm Image',output)

cv.waitKey(0)

 

 

 

Output

2.Write a function that takes dermoscopic skin image (used in task 3 as well) as input and segments the image using K-mean clustering with and k=2. Display the resulting image.

Code

import numpy as np
import cv2 as cv
img = cv.imread('freestudystrom.bmp',0) # Read input image
output1 = np.copy(img)       #image for mean thresholding/segmentation
output2 = np.copy(img)       #image for binary segmentation 
dims = np.shape(img)         #image dimensions
rows = dims[0]               #rows of image
cols = dims[1]               #cols of image
M1 = np.min(img) + 10        #Initial Mean for Cluster 1
M2 = np.max(img) - 20        #Initial Mean for Cluster 2

Cluster1 = []                #Cluster 1 defined
Cluster2 = []                #Cluster 2 defined

while(1):
    for i in range(0, rows):
        for j in range(0, cols):
            diff1 = abs(img[i, j] - M1)         #difference of pixel value and Mean 1
            diff2 = abs(img[i, j] - M2)         #difference of pixel value and Mean 2
            decide = min(diff1, diff2)          #Comparative minimum difference
            if decide == diff1:                 #Closer to Mean 1
                Cluster1.append(img[i, j])      #Add to Cluster 1
            else:                               #Closer to Mean 2
                Cluster2.append(img[i, j])      #Add to Cluster 2
    N1 = np.mean(Cluster1)                      #Calculate mean for Cluster 1
    N2 = np.mean(Cluster2)                      #Calculate mean for Cluster 2
    if (N1==M1) and (N2==M2):                   #End when means become consistent
        break
    else:                                       #Update Means and Make Clusters empty again
        M1 = N1
        M2 = N2
        Cluster1 = []
        Cluster2 = []

L1 = len(Cluster1)                              #length of Cluster 1
L2 = len(Cluster2)                              #length of Cluster 2

for k in range(0, L1):                         
    output1[output1 == Cluster1[k]] = N1        #assign Mean 1 to Cluster 1 pixels
for k in range(0, L2):
    output1[output1 == Cluster2[k]] = N2        #assign Mean 2 to Cluster 2 pixels
   
for k in range(0, L1):
    output2[output2 == Cluster1[k]] = 0         #assign 0 to Cluster 1 pixels
for k in range(0, L2):
    output2[output2 == Cluster2[k]] = 255       #assign 255 to Cluster 1 pixels

cv.imshow('Mean Thresholded Image',output1)
cv.imshow('Final Binary Image',output2)
cv.waitKey(0)

 

Output