EEE-6512_Exam3_Lecture16

What is segmentation

Segmentation divides an image into groups of pixels

Why are pixels grouped together in segmentation

They share some local property such as gray level or color or texture or motion

What are four ways of displaying segmentation output shown in the lecture

Boundaries and labels and pseudocolors and mean colors

What is oversegmentation

Dividing what should be one group into too many groups

What is undersegmentation

Failing to separate regions that should be distinct

What is the simplest segmentation problem discussed in the lecture

Foreground background segmentation

What is the goal of foreground background segmentation

Separate foreground objects from background

What common approach is used for the simplest segmentation problem

Find a threshold value that separates foreground pixels from background pixels

According to the lecture what do points represent in segmentation context

Foreground objects

According to the lecture what do edges represent in segmentation context

Boundaries between background and foreground

In global thresholding when is g(x y)=1

When f(x y) is greater than T

In global thresholding when is g(x y)=0

When f(x y) is less than or equal to T

What is multiple thresholding used for

Partitioning intensities into more than two classes

How can noise affect image thresholding according to the lecture

It can blur or overlap histogram modes and make separation harder

How can illumination and reflectance affect thresholding

They can distort the histogram and make class separation less clear

What is Step 1 of the basic global thresholding algorithm

Select an initial estimate for the global threshold T

What is Step 2 of the basic global thresholding algorithm

Segment the image using T into G1 and G2

What is G1 in the basic global thresholding algorithm

The group of pixels with intensity values greater than T

What is G2 in the basic global thresholding algorithm

The group of pixels with intensity values less than or equal to T

What is Step 3 of the basic global thresholding algorithm

Compute the average intensity values m1 and m2 for G1 and G2

What is Step 4 of the basic global thresholding algorithm

Compute a new threshold T as one half of m1 plus m2

What is Step 5 of the basic global thresholding algorithm

Repeat the process until the threshold change between iterations is smaller than delta T

When does the basic global thresholding algorithm work well

When there is a reasonably clear valley between histogram modes for objects and background

Under what condition does the basic global thresholding algorithm converge in a finite number of steps

When the initial T is greater than the minimum and less than the maximum intensity value

What is the basic idea of Otsu's method

The best threshold is the one that gives the best separation between classes in intensity

Where are all operations performed in Otsu's method

Directly on the histogram

What does Otsu's method assume about the histogram and image

It assumes the histogram and image are bimodal

What does Otsu's method assume about statistics in the image

It assumes stationary statistics

What illumination assumption is made implicitly in Otsu's method

It assumes uniform illumination

Why does Otsu's method implicitly assume uniform illumination

So bimodal brightness behavior arises from object appearance only

What search strategy does Otsu's method use to find the threshold

An exhaustive search for the optimal threshold

What quantity does Otsu's method maximize

The between class variance

What is k* in Otsu's method

The threshold value that maximizes the between class variance

What does the separability measure eta equal in Otsu's method

Between class variance divided by global variance

What is Step 1 of Otsu's algorithm summary

Compute the normalized histogram of the input image

What is Step 2 of Otsu's algorithm summary

Compute the cumulative sums P1(k)

What is Step 3 of Otsu's algorithm summary

Compute the cumulative means m(k)

What is Step 4 of Otsu's algorithm summary

Compute the global intensity mean mG

What is Step 5 of Otsu's algorithm summary

Compute the between class variance for each k

What is Step 6 of Otsu's algorithm summary

Obtain the Otsu threshold k*

What is Step 7 of Otsu's algorithm summary

Obtain the separability measure

Why can image smoothing improve global thresholding

Smoothing reduces noise and can make histogram classes more separable

What problem is raised in the lecture when the foreground size is significantly reduced

The histogram may be dominated by the background and thresholding may become less reliable

What is the purpose of using edges to improve global thresholding

To build a histogram from more informative pixels so global segmentation works better

What is Step 1 of using edges to improve global thresholding

Compute an edge image using either gradient magnitude or the absolute value of the Laplacian

What is Step 2 of using edges to improve global thresholding

Specify a threshold value T

What is Step 3 of using edges to improve global thresholding

Threshold the edge image to produce a binary mask and select pixels from f(x y)

What is Step 4 of using edges to improve global thresholding

Compute a histogram using only the chosen pixels from f(x y)

What is Step 5 of using edges to improve global thresholding

Use that histogram to segment f(x y) globally

What two edge based images can be used in Step 1 of edge improved global thresholding

Gradient magnitude or absolute Laplacian

In the edge improved thresholding method what is the role of the binary image from Step 3

It acts as a mask for selecting pixels from the original image

Why does the edge improved thresholding method compute a histogram from only chosen pixels

Because those pixels can give a more useful histogram for separating classes

What is the main difference between basic global thresholding and Otsu's method

Basic global thresholding iteratively updates T using class means while Otsu's method searches for the threshold that maximizes between class variance

What is the main difference between standard Otsu thresholding and edge improved thresholding

Standard Otsu uses the full image histogram while the edge improved version uses a histogram built from selected pixels

What is the goal of multiple thresholds

To divide the image into more than two classes

In multiple threshold Otsu style segmentation what is optimized

The between class variance across multiple classes

What does the iceberg example illustrate in the lecture

That dual thresholds can segment an image into three regions

How does Lecture 16 define segmentation at a high level

It divides an image into groups of pixels based on shared local properties

What common segmentation mistakes shown in the lecture should you recognize

Oversegmentation and undersegmentation

What kind of professor style question is likely from this lecture about Otsu's method

Questions on assumptions and optimization target and algorithm steps

What kind of professor style question is likely from this lecture about thresholding workflow

Questions on exact step order and what each group or mask represents