CV Answer Keys

1. Image Degradation/Restoration Model:

Imagine taking a photograph through a foggy window or with a shaky camera. The resulting image is
not a perfect representation of the scene—it’s degraded. In digital image processing, we model this
degradation using a mathematical formula so we can attempt to undo it and restore the original image.

The model is:

g(x, y) = h(x, y) ∗ f(x, y) + η(x, y)

Let’s break this down:

• f(x, y): The original image (what we want to recover).

• h(x, y): The degradation function. This represents things like blurring from motion or an out-of-
focus lens. Think of it as a "smudge" function that distorts the image.

• ∗: Convolution, which means the original image is combined with the degradation function.

• η(x, y): Additive noise—random variations caused by things like electronic sensor
imperfections, transmission interference, or atmospheric conditions.

• g(x, y): The final degraded image—the one we actually observe.

The goal of image restoration is to use knowledge or estimates of h(x, y) and η(x, y) to reconstruct an
approximation of f(x, y). In other words, we want to remove or reverse the effects of both the blur and
the noise to retrieve a clean image.

In the frequency domain (which deals with the image in terms of its frequency components rather than
pixel intensities), this becomes:

G(u, v) = H(u, v) · F(u, v) + N(u, v)

Here, the capital letters represent the Fourier transforms of the respective spatial domain functions. This
frequency-domain representation is particularly useful because convolution in the spatial domain
becomes multiplication in the frequency domain, which is easier to handle computationally.

Key takeaway: The more accurately we know h(x, y) and η(x, y), the better we can estimate f(x, y). This
process is objective and based on modelling physical causes of degradation.

2. Convert RGB to HSI model

To convert a pixel from RGB to HSI, where R, G, B ∈ [0, 1], the following steps are used:

Step 1: Normalize R, G, B values to the range [0, 1] if they aren't already.

Step 2: Compute Intensity (I):

I = (R + G + B) / 3

Step 3: Compute Saturation (S):

 This measures the degree to which a colour is diluted with white.

Step 4: Compute Hue (H):

Use the intermediate variable:

Then,

If B ≤ G: H = θ
If B > G: H = 360° − θ

Convert H to range [0,1] by dividing by 360 if needed.

Step 5: HSI values are then:

• H ∈ [0,1] (or in degrees 0° to 360°)

• S ∈ [0,1]

• I ∈ [0,1]

Note:

• If R = G = B, the colour is Gray (S = 0), and H is undefined or set to 0.

• These computations are performed on a per-pixel basis.

3. Explain Edge Detection and Gradient-Based Method:

Edge detection is a fundamental image segmentation technique used to identify points in a digital image
where brightness changes sharply. These points typically represent object boundaries.

Gradient-Based Edge Detection:

This method is based on detecting sharp changes in intensity by computing the gradient of the image.
The key ideas are:

a) Image Gradient:
The gradient of an image f(x, y) at a point gives the direction and rate of fastest increase in
intensity. It is a vector:

∇f(x, y) = [∂f/∂x, ∂f/∂y]

b) Gradient Magnitude and Direction:

The strength (magnitude) of the edge is given by:
 M(x, y) = √[(∂f/∂x)² + (∂f/∂y)²]
The direction is given by:

α(x, y) = arctangent(∂f/∂y / ∂f/∂x)

These values indicate how strong and in which direction the intensity changes, which helps detect
edges.

c) Operators for Gradient Approximation:

Several filters are used to approximate derivatives:

• Sobel Operator: Emphasizes central pixels, smooths noise.

• Prewitt Operator: Similar to Sobel but simpler.

• Roberts Operator: Uses 2×2 diagonal filters.

• Kirsch Operator: Emphasizes edges in particular compass directions.

d) Smoothing and Thresholding:
To reduce noise before gradient calculation, smoothing (e.g., averaging filters) is applied. After
computing the gradient, thresholding is used to suppress weak edges (usually caused by noise),
enhancing strong, meaningful edges.

Summary:
Gradient-based edge detection detects intensity changes using the first derivative of the image. It’s
sensitive to noise, so smoothing and thresholding are commonly used. The gradient magnitude
shows edge strength, and its direction indicates orientation.

4. Explain Hit-Miss transform

The Hit-or-Miss Transform (HMT) is a morphological operation used primarily for shape detection in
binary images. Unlike basic morphological operations like dilation or erosion, HMT utilizes two
structuring elements: one to probe the foreground (object) and one to probe the background.

Definition:

Let I be the binary image composed of:

• Foreground: A

• Background: Ac (complement of A)
Let B1 be a structuring element for detecting a specific foreground pattern, and B2 for detecting a
background pattern.

The HMT is defined as:

I ⊛ (B1, B2) = (A ⊖ B1) ∩ (Ac ⊖ B2)

This means:

• B1 must fit within the foreground (A)

• B2 must simultaneously fit within the background (Ac)

 • The transform identifies locations where both conditions are true.
Use Case:

• Detecting specific shapes or patterns (like corners, holes, or particular object outlines) in binary
images.

Example:

If you want to find a 3×3 cross shape within an image, you can define B1 to match the cross in the
foreground and B2 to match the surrounding background. The intersection of the two erosions pinpoints
where the pattern occurs.

Advantages:

• Highly selective; useful for exact shape detection.

• Can differentiate between similar structures based on foreground and background geometry.

5. Difference between RGB and CMY Models

Feature RGB Colour Model CMY Colour Model

Type Additive colour model Subtractive colour model

Primary Components Red, Green, Blue Cyan, Magenta, Yellow

Colour Generation Colours are created by adding light of Colours are created by subtracting light
Method R, G, B from white

Basis Used in devices that emit light (e.g., Used in devices that reflect light (e.g.,
monitors, cameras) printers, copiers)
Secondary Colours Cyan (G + B), Magenta (R + B), Red (M + Y), Green (C + Y), Blue (C + M)
Yellow (R + G)

Black Representation (0, 0, 0) — absence of light (1, 1, 1) — full absorption of R, G, B

White Representation (1, 1, 1) — full presence of light (0, 0, 0) — no absorption; full reflection of
R, G, B

Mathematical Relation RGB → CMY: C = 1 – R, M = 1 – G, CMY → RGB: R = 1 – C, G = 1 – M, B =

Y=1–B 1–Y

Application Used in displays, scanners, digital Used in colour printing and inkjet devices
cameras

2. Where They Are Used:

• RGB Model:

o Used in hardware that emits light, such as:

▪ Computer monitors
 ▪ Televisions

▪ Digital cameras

▪ Smartphones

o It is the standard model for image acquisition and display systems.

o Ideal for image generation and screen-based applications.

• CMY (and CMYK) Model:

o Used in colour reproduction systems that use pigments or dyes, such as:

▪ Inkjet printers

▪ Photocopiers

▪ Press printing (CMYK adds black (K) to compensate for imperfect colour inks)

o CMY is derived from RGB and is used for converting screen colours to printable output.