Section - A

All parts of question-1 are compulsory. However, internal choices are given. Each part carries ONE

mark. Answer each part in about 45-60 words.

[3X1=3]

Q1.

(a) Define depth of field.

- The depth of field (DoF) refers to the range of distances within a scene that appear acceptably

sharp in an image. It depends on factors like aperture size, focal length, and distance between the

camera and the subject. A smaller aperture increases the depth of field, making more of the scene

appear sharp.

(b) How does the size of the pinhole affect the image quality?

- The size of the pinhole in a pinhole camera affects image sharpness and brightness.

- A smaller pinhole improves sharpness by reducing light scattering but may make the image

dimmer.

- A larger pinhole allows more light, making the image brighter but less sharp due to increased

diffraction.

OR
(b) Consider a 3-bit gray scale image with dimensions 32 × 32. What will be the range of values in

its X-axis?

- For a 3-bit grayscale image, pixel intensity values range from 0 to 7 (since 2^3 = 8).

- The X-axis in a 32 × 32 image will span 0 to 31, corresponding to the column indices.

(c) Give any three applications of computer vision.

1. Object detection: Used in autonomous vehicles to identify objects like pedestrians and traffic

signs.

2. Facial recognition: Applied in security systems for identity verification.

3. Medical imaging: Enhances diagnostic accuracy by analyzing X-rays, MRIs, and CT scans.

---

Section - B

All questions are compulsory. However, internal choices are given. Each question carries THREE

marks. Answer each question in about 130-180 words.

[4 X 3 =12]

Q2. Describe the types of 2D transformations commonly used in image processing.

1. Translation: Shifts an image by moving every pixel a fixed distance in a specified direction.

- Represented by:

[x'] = [x + tx]
 [y'] [y + ty]

2. Rotation: Rotates an image around a point (usually the origin) by a specified angle (theta).

- Represented by:

[x'] = [costheta -sintheta] [x]

[y'] [sintheta costheta] [y]

3. Scaling: Changes the size of an image by multiplying pixel coordinates by scaling factors (sx, sy).

- Represented by:

[x'] = [sx 0 ] [x]

[y'] [ 0 sy] [y]

4. Shearing: Tilts an image along the X or Y axis.

---

Q3. Differentiate between low pass and high pass filters.

| Aspect | Low Pass Filter | High Pass Filter |

|---------------------|--------------------------------------------|-------------------------------------------|

| Function | Passes low-frequency components, blocks high-frequency. | Passes

high-frequency components, blocks low-frequency. |

| Effect on Image | Smoothens the image, reducing noise and details. | Enhances edges and fine

details. |

| Example | Gaussian or Average filter. | Sobel, Prewitt, or Laplacian filter. |

OR
Q3. How are edges detected using the Laplacian of Gaussian (LoG) method?

1. Smoothing: The image is smoothed using a Gaussian filter to reduce noise.

2. Second Derivative: The Laplacian operator (second derivative) is applied to identify regions of

rapid intensity change.

3. Zero-Crossings: Edges are detected by finding zero-crossings in the Laplacian response.

4. Implementation: LoG combines smoothing and edge detection into a single operation using a

convolution kernel.

---

Q4. Examine how a higher stride value contributes to the downsampling of an image and suggest

methods to address this issue.

Effect of Stride:

- A higher stride reduces the overlap between the receptive fields of a convolution operation,

effectively downsampling the image.

- While this decreases computational cost, it may result in loss of important spatial details.

Methods to Address:

1. Smaller Stride: Use a stride of 1 or 2 to retain more details.

2. Padding: Apply zero-padding to preserve edge information.

3. Dilated Convolutions: Expand the receptive field without increasing stride.

Q5. Prewitt operator on highlighted pixel (2,3):

Image:

[0 1 4 3]

[2 4 3 0]

[0 1 2* 7]

[4 1 5 6]

Vertical Prewitt kernel:

[-1 0 1]

[-1 0 1]

[-1 0 1]

Horizontal Prewitt kernel:

[ 1 1 1]

[ 0 0 0]

[-1 -1 -1]

Apply kernels and calculate Gx and Gy, then find G = sqrt(Gx^2 + Gy^2).

---

Section - C
Question-6 is compulsory. However, internal choices are given. Each question carries Five marks.

Answer each question in about 225-300 words.

[1X5=5]

Q6. Algorithm for detecting lines using Hough Transform:

1. Edge Detection: Apply an edge detection algorithm (e.g., Canny) to extract edges from the image.

2. Hough Space: Define a parameter space (rho, theta) where lines are represented.

3. Voting: For each edge pixel, calculate rho = x costheta + y sintheta for various theta values and

increment the corresponding accumulator cell.

4. Peak Detection: Identify peaks in the accumulator array as potential lines.

5. Line Reconstruction: Map the detected peaks back to the image space to draw the lines.

OR

Q6. Double thresholding and hysteresis in edge detection:

Double Thresholding:

- Uses two thresholds (Tlow and Thigh) to classify edges.

- Pixels with gradient magnitude above Thigh are strong edges, while those between Tlow and

Thigh are weak edges.

Hysteresis:

- Links weak edges to strong edges if they are connected, ensuring continuity.
- Weak edges not connected to strong edges are suppressed as noise.

These steps ensure accurate and robust edge detection.