CSL7320: Digital Image

Analysis
Segmentation
Image Segmentation

Slide credit: Yan Tong

Image Segmentation - Applications

Slide credit: Yan Tong

Brief Review of Connectivity

Slide credit: Yan Tong

Image Segmentation

Slide credit: Yan Tong

Image Segmentation

Slide credit: Yan Tong

Point, Line and Edge Detection

Slide credit: Yan Tong

Background

Slide credit: Yan Tong

Some Observations

Slide credit: Yan Tong

Detection of Isolated Points

*
*

Slide credit: Yan Tong

Line Detections by Laplacian

Slide credit: Yan Tong

Line Detections In Specified Directions

Slide credit: Yan Tong

An Example

Slide credit: Yan Tong

Edge Models

Slide credit: Yan Tong

Edge Models

Slide credit: Yan Tong

Derivatives – Idea Cases

whether an edge pixel lies on the dark or light side of an edge

produces two values for every edge in an image
/thick edges Slide credit: Yan Tong
 But, In Practice …

In summary, the three steps performed

typically for edge detection are:
1. Image smoothing for noise reduction. The
need for this step is illustrated by the results in
the second and third columns of Fig. 10.11.
2. Detection of edge points. As mentioned
earlier, this is a local operation that extracts
from an image all points that are potential
edge-point candidates.
3. Edge localization. The objective of this step
is to select from the candidate points only the
points that are members of the set of points
comprising an edge. Slide credit: Yan Tong
Basic Edge Detection

Slide credit: Yan Tong

Masks for Calculating the Gradient (2x2)

Along x => g_x

Along y => g_y

Along x => g_x

Along y => g_y

Slide credit: Yan Tong

Masks for Calculating the Gradient (3x3)

Along y => g_y

Along x => g_x

Slide credit: Yan Tong

An Example
Along x => g_x Along y => g_y

Slide credit: Yan Tong

An Example

Angle information is employed in Canny edge detector and other feature

representation, such as Histogram of Orientated Gradients (HOG). Slide credit: Yan Tong
An Example

Slide credit: Yan Tong

Combining the Gradient with Thresholding:
An Example

Slide credit: Yan Tong

Edge Magnitude and Direction (Angle)
The Kirsch compass kernels in Fig. 10.15, are designed to detect edge magnitude and
direction (angle) in all eight compass directions.
0 degree
Edge Magnitude and Direction (Angle):
Example
Summary on Simple Edge Detectors

Slide credit: Yan Tong

Advanced Edge Detection Techniques

Slide credit: Yan Tong

Marr-Hildreth Detector (LoG)

Slide credit: Yan Tong

Marr-Hildreth Detector (LoG)

Slide credit: Yan Tong

Marr-Hildreth Detector (LoG)

Slide credit: Yan Tong

LoG Filtering

Slide credit: Yan Tong

An Example – Edges are 1 Pixel Thick

Slide credit: Yan Tong

Approximate LoG by DoG

Slide credit: Yan Tong

Canny Edge Detector

Slide credit: Yan Tong

Canny Edge Detector

Slide credit: Yan Tong

Quantize the Edge Direction

Slide credit: Yan Tong

Canny Detector -- Algorithm

+ve, -ve side

If we set the threshold too

low, there will still be some
false edges (called false
positives). If the threshold is
set too high, then valid edge
points will be eliminated
(false negatives).

Slide credit: Yan Tong

Canny Detector -- Algorithm
Double Thresholding: Example Two
Uses two thresholds: a low threshold, T_L and a high threshold, T_H. thresholded
images

Modified
Canny Detector -- Algorithm
Double Thresholding: Example
Canny Detector -- Algorithm
Double Thresholding
An Example

Slide credit: Yan Tong

An Example

Slide credit: Yan Tong

Edge Linking and Boundary Detection

Global Processing

Slide credit: Yan Tong

Edge Linking – Local Processing

where E is a positive threshold.

Similarly, where A is a positive angle threshold

Slide credit: Yan Tong
Edge Linking – Local Processing
Simplified Procedure
Edge Linking – Local Processing Example

Slide credit: Yan Tong

Edge Linking – Hough Transform
Global Processing

O(n^2)

O(n^3)

Slide credit: Yan Tong

Hough Transform

Slide credit: Yan Tong

Hough Transform

Slide credit: Yan Tong

Hough Transform: Toy example

A: green line
B: red line

Slide credit: Yan Tong

Hough Transform: Real example

Slide credit: Yan Tong

Hough Transform: Real example

Slide credit: Yan Tong

Notes

Slide credit: Yan Tong

Intensity Thresholding

Slide credit: Yan Tong

Key Factors Affect Thresholding

Slide credit: Yan Tong

The Role of Noise in Image Thresholding

Slide credit: Yan Tong

The Role of Illumination in Thresholding

Slide credit: Yan Tong

How to Pick the Threshold

Slide credit: Yan Tong

An Example

Slide credit: Yan Tong

Optimal global thresholding using Otsu’s
method

Slide credit: Milton Maritz

Optimal global thresholding using Otsu’s
method

Slide credit: Milton Maritz

Optimal global thresholding using Otsu’s
method

Slide credit: Milton Maritz

Optimal global thresholding using Otsu’s
method

Slide credit: Milton Maritz

Optimal global thresholding using Otsu’s
method

Slide credit: Milton Maritz

Optimal global thresholding using Otsu’s
method
Optimal global thresholding using Otsu’s method

Slide credit: Milton Maritz

Optimal global thresholding using Otsu’s
method

Slide credit: Milton Maritz

Optimal global thresholding using Otsu’s
method

Slide credit: Milton Maritz

Multiple Thresholds

Slide credit: Milton Maritz

Multiple Thresholds

Slide credit: Milton Maritz

Multiple Thresholds

Slide credit: Milton Maritz

Variable Thresholding

Slide credit: Milton Maritz

Variable Thresholding

Slide credit: Milton Maritz

Variable Thresholding

Slide credit: Milton Maritz

Region-Based Segmentation

Slide credit: Yan Tong

Region Growing Algorithm

Slide credit: Yan Tong

Region Growing Algorithm

Slide credit: Milton Maritz

An Example

Slide credit: Yan Tong

Region-Splitting and Merging Algorithm

Slide credit: Yan Tong

Region-Splitting and Merging Algorithm

Slide credit: Milton Maritz

An Example

Slide credit: Yan Tong

Advanced Approaches for Image
Segmentation

http://blog.qure.ai/notes/semantic-segmentation-deep-learning-review

Slide credit: Yan Tong

Image Segmentation: K-Means Clustering
Image Segmentation: K-Means Clustering
Additional Notes and Readings

Click here

Click here

Click here
Slide credit: Yan Tong