Indian Institute Of Technology-Image Segmentation

Overview

• What is segmentation?

• Different approaches for image segmentation

  • Discontinuity based

  • Similarity based

• Different edge detection operators

• Linking of edge points

  • Local processing

  • Global processing

Image Segmentation two Approaches

1. Discontinuity Based

   1. Isolated Points

   2. Lines

   3. Edges

2. Similarity Based

   1. Thresholding

   2. Region Growing

   3. Region Splitting & Merging

We are going to focus on Discontinuity Base Image Segmentation

Point Detection ( $|R| > T$ ) (T = non-negative threshold)

Line Detection (masks = [horizontal, vertical, 45, -45])

Edge Detection

Don't forget: using $2^{nd}$ derivative might look pretty sensible, however, it has a huge downside which is sensivity to noise in order to fix this problem we usually use GaussianBlur. Stay tuned!

Calculation of Derivative

Calculation of Direction

Edge Operators

Prewitt Edge Operator

Sobel Edge Operator

Example of Sobel Edge Operator Result

1. Original Image

2. Horizantal Component of Sobel Edge Detector

3. Vertical Component of Sobel Edge Detector

4. Combining Vertical and Horizontal Sobel EDge Detector

Laplacian Operator

Laplacian Operator horizontal and Vertical

If we also want to add diagonal elements, laplacian operator transforms like

Laplacian Of Gaussians (LoG)

Why we need LoG, Using Laplacian what kind of downsides have? These Questions are critically important to understand the Log. We know that Laplacian uses 2nd derivative function and it's really really sensitive to noise that's a big problem but thanks to gaussian we can solve this problem and that's why we use gaussian mask before implementing laplacian mask.

Laplacian of Gaussian Graphics

1. Implemented Gaussian

2. After Gaussian implementing Laplacian

LoG Mask

Comparison Result Images

1. Original Image

2. Output of the Sobel Operator

3. Output of the LoG operator

LoG(Laplacian of Gaussian) can determine that what is the location of an edge presenting on the image.

We learned that normally we don't use second derivative of function on edge detection, however, important to learn LoG operator.

Edge Linking

1. Local Processing

2. Global Processing

We are going to look at local processing technique

Suppose that image already operated by sobel edge operator.

In this sobel edge operated image we are going to look for similarities.

Similarity Measures.

• Strength

• Direction (of gradient)

T = Thresholding

A = Angle

Quiz Questions

My Answers

1. Image segmentation is simply distinguishing an object from rest of the image.

2. Basic approaches are Discontinuity based method (Edge Based Method), Region Based Method(Similarity Based). These were basic methods and there are some advance methods as well, I am going to add an image for that at the last of the page.

3. Line and Edge different because in line background is the same but there is a line top of the background. However, in edge there is a transmisson between two area.

4. In Prewitt operator there is no any emphasis but in Sobel emphasis increases when we close to the center of an image.

5. LoG stands for Laplacian of Gaussian operation. In LoG we simply apply Gaussian operation then Laplacian operator with this way we're going to be able to obtain an awesome result.

All Image Segmentation Methods

Comparison Of Image Segmentation Methods

Extra Source: