Image Segmentation using MATLAB Digital Images Processing Computer Vision python openCV machine learning deep learning

Image segmentation Using MATLAB
Digital Images Processing Using MATLAB

What is segmentation?

Image segmentation is the process of partitioning an image into a collection of connected sets of pixels. The goal of segmentation is to simplify change the representation of an image into something that is more meaningful and easier to analyses. Image segmentation is typically used to locate objects and boundaries like lines, curves in images. It is the process of assigning a label to every pixel in an image. 

Three main techniques to do.

Thresholding

Region based segmentation

Edge based segmentation

Thresholding

Thresholding is finding histogram of gray level intensity.

Types

Basic Global Thresholding

Multiple Threshold

Variable Thresholding

Otsu’s Technique

Basic Global Thresholding

Primarily Segment image use:

Calculate the average intensity m1 and m2 for the pixels

Calculate a fresh threshold:       

Until the variance between values of T is minor than a predefined parameter.

Otsu’s Technique

Based on a very simple idea: Find the threshold that minimizes the weighted within-class variance. 

This turns out to be the same as maximizing the between-class variance.

Operates directly on the gray level histogram.

Multiple Threshold

As Otsu’s technique, it takes extra area and k* 

Disadvantage:

It becomes too complex when amount of region more than two or three.  

Variable Thresholding

Image partitioning

It is work once the objects of interest and the background inhabit areas of sensibly similar size. If not, it will fail.

Variable thresholding based on local image properties

By moving average

It debated is based on computing a moving average along scan appearances of an image.

Region based segmentation

Region Segmentation is procedure of finding region, but not finding edge

Region Growing

Cheng-Jin Kuo`s method is used

Data Grouping (Clustering Technique)

Partitional clustering

Algorithm:

Select a random pixels

Usage 8-connected and threshold to decide

Repeat a and b until nearly points are classified.

Reproduction of region growing (90% pixels )

Threshold/second: 20/4.7 seconds.

Data Clustering

Using centroid to signify the huge amounts of clusters

Partitional clustering, we have to choose the number of clustering we want first before we begin the process.

Hierarchical clustering, we can alteration the number of cluster anytime throughout process if we need.

Hierarchical clustering

Algorithm of hierarchical accumulation (built)

Understand every solo data as a cluster Ci

Find out Ci, Cj for the distance is the shortest.

Repeat the steps until satisfies our request.

d(a,b) as the space between data a and b

Algorithm of hierarchical division (break up )：

 Diameter of cluster  

Partitional clustering

Choose the numbers of the cluster (k-means)

ADVANTAGES & DISADVANTAGES

Hierarchical algorithm

Advantages

Result is reliable

Idea is simple

Disadvantages

It is consuming, so is not suitable for a big database.

Partitional algorithms

Advantages

Numbers of cluster is fixed, so the concept is also simple.

Calculating speed is fast.

Disadvantages 

Determine the number of clusters.

Initial problem

Edge based segmentation

Edge-Based Segmentation

Edge-Based Segmentation is by mask to notice edge in image by convolution.

Basic Edge Detection

SRHLT - Short response Hilbert transform

Watersheds

The Marr Hildreth edge detector (LoG)

Basic Edge Detection                

Gradient

Gradient is to find edge strength and direction at site (x,y) of image.

The magnitude of vector, meant as M(x,y)

The direction of the gradient vector is assumed by the angle

Roberts 

Robert edge detection, the perpendicular and horizontal edges carry out separately and then place together for resulting edge detection. The Roberts edge detector uses the next masks to approximate digitally the first derivatives as differences between adjacent pixels.  

Prewitt operator

Prewitt operator edge detection masks are the one of the oldest and greatest understood methods of detecting edges in images The Prewitt edge detector uses the following mask to approximate digitally the first derivatives Gx and Gy.

Sobel operator

The sobel edge detector calculates the gradient by using the discrete differences between rows and columns of a 3X3 neighborhood. The sobel operator is based on convolving the image with a minor, divisible, and number valued filter.

Canny Edge Detection

Canny edge detection is a multistage algorithm to detect a extensive range of edges in images. This detector finds edges by observing for local maxima of the gradient off (x, y). The gradient is calculated using the derivative of a Gaussian filter. The technique uses two thresholds to detect strong and weak edges and includes the weak edges in the output only if they are connected to strong edges

The Marr-Hildreth edge detector (LoG)

This is second-order deviation, known as Laplacian.

Filter the input image with an n*n Gaussian low pass filter. 99.7% of the capacity under a 2-D Gaussian surface lies between around the mean.

Short response Hilbert Transform (SRHLT)

Hilbert transform

Short response Hilbert transform (SRHLT)

Watersheds

Algorithm:

Markers

Outside markers:

Points along the watershed line along maximum points.

Interior markers: 

All points in connected component have the similar intensity.

Points in area form a connected component 

That is surrounded higher points.

MATLAB Source Code For Image Segmentation

Flow Chart 

%IMAGE SEGMENTATION

clc;

clear all;

close all;

%edge detection

a=imread('autumn.tif');

b=rgb2gray(a);

figure(1);

imshow(b);

title('original image');

%roberts operator

c=edge(b,'roberts');

figure(2);

imshow(c);

title('roberts image');

%sobel operator

d=edge(b,'sobel');

figure(3);

imshow(d);

title('sobel image');

%prewitt operator

e=edge(b,'prewitt');

figure(4);

imshow(e);

title('prewitt image');

%log operator

f=edge(b,'log');

figure(5);

imshow(f);

title('log image');

%canny operator

g=edge(b,'canny');

figure(6);

imshow(g);

title('canny image');

%point detection

h=[0,1,0;1,-4,1;0,1,0];

i=imfilter(b,h,'symmetric','conv');

figure(7);

imshow(i);

title('laplacian point filtered image');

%line detection

j=im2bw(b);

w=(1/3)*[1,0,-1;1,0,-1;1,0,-1];

k=imfilter(double(j),w);

k(k<1)=0;

figure(8);

imshow(k);

title('line detection image');

Output for Image Segmentation

Fig.1 Original Image

Fig.2. Gray Scale image

Fig.3. Robert image

Fig.4. Sobel Image

Fig.5. Prewitt Image

Fig.6. Log image

Fig.7. Canny Image

Fig.8. Laplacian Point Filtering Image

Fig.9. line detection Image

Application

Real-world applications of image segmentation are:

Machine vision

Deep Learning

Medical imaging, including volume rendered images from CT (Computed Tomography) and MRI (Magnetic Resonance Imaging).

Measure tissue volumes

Surgery planning

Intra-surgery navigation

Locate tumors and pathologies 

Virtual surgery simulation

Diagnosis, learning of anatomical structure 

Pedestrian recognition

Object recognition

Face detection

Face recognition

Recognition Tasks

Fingerprint recognition

Iris recognition

Traffic controller schemes

Brake light detection

Video investigation

Find objects in satellite images like roads, forests, crops.

Conclusion

Since edge detection is the first step in object boundary extraction and object recognition, it is significant to know the differences between different edge detection operators. from this section an attempt is made to review the edge detection methods which are based on discontinuity intensity levels. The relative presentation of various edge detection techniques is carried out with two images by using MATLAB. It have been experiential that that the Canny edge detector produces greater accuracy in detection of object edges with greater entropy, PSNR, MSE and  implementation time compared with Sobel, Roberts, Prewitt, Zero crossing and LOG.