1. What is Computer Vision?
Computer Vision is a field of Artificial Intelligence that teaches computers to see, understand, and interpret images and videos like humans do.
Example: Recognizing faces, reading signs, or detecting objects in a picture.
2. How does Computer Vision work?
1. Capturing images using cameras or sensors.
2. Preprocessing the images to remove noise and enhance quality.
3. Extracting features like edges, shapes, and colors.
4. Detecting and segmenting objects in the image.
5. High level processing where AI models classify, recognize, or make decisions based on the image.
3. What is a pixel?
A pixel (picture element) is the smallest unit of an image. Each pixel has a color or brightness value, and many pixels together form the full image.
Example: An image with 1920×1080 resolution has 1920 columns and 1080 rows of pixels.
4. What is resolution?
Resolution refers to the number of pixels in an image. It is written as width × height (e.g., 1920×1080). Higher resolution means more pixels, better clarity, and more details.
5. What is bit depth?
Bit depth is the number of bits used to represent the color of a single pixel.
1 bit: Black and White (2 colors).
8 bit: 256 shades of gray.
24 bit: 16.7 million colors (RGB: 8 bits each for Red, Green, Blue).
Higher bit depth means more color accuracy.
6. Explain image file formats JPEG, PNG, and TIFF with features.
JPEG (Joint Photographic Experts Group):
- Compressed format (lossy).
- Small file size, good for photos.
- Loses some quality when compressed.
PNG (Portable Network Graphics):
- Lossless compression.
- Supports transparency.
- Larger file size than JPEG
- Good for graphics and logos.
TIFF (Tagged Image File Format):
- Very high quality
- Large file size.
- Lossless format, used in medical/scientific imaging.
- Supports multiple layers and high bit depth.
7. How are images represented in numbers using RGB and Grayscale models?
RGB Model: Each pixel is represented using three values (Red, Green, Blue), each ranging from 0–255.
Example: (255, 0, 0) represents pure Red. (0,255, 0) represents pure Green and Example: (0, 0, 255) represents pure Blue.
Grayscale Model: Each pixel is represented by a single value (0–255) showing brightness. 0 = black, 255 = white.
8. What are the components of the Computer Vision process?
- Image Acquisition: Capturing an image using a camera or sensor.
- Preprocessing: Enhancing image quality, reducing noise, resizing.
- Feature Extraction: Identifying important features (edges, shapes, textures).
- Detection and Segmentation: Locating and separating objects of interest.
- HighLevel Processing: Interpreting results, classification, decision-making (e.g., face recognition, object detection).
9. What happens during Preprocessing of image?
- Noise Reduction
- Image Normalization
- Resizing/Cropping
- Histogram Equalization
10. What is Noise Reduction?
Noise Reduction is a process that removes unwanted spots, grain, or distortions from an image to make it clearer and easier to analyze.
11. What is Image Normalization?
Image Normalization is the process of adjusting the brightness or color values of an image to a standard range so it is consistent and easier for a computer to process.
12. What is Resizing Images?
Resizing Images means changing the width and height of an image to make it bigger or smaller.
13. What is Cropping Images?
Cropping removes unwanted parts of an image. Example: Cropping a face from a group photo for face recognition.
14. What is Histogram Equalization?
Histogram Equalization is a method to improve the brightness and contrast of an image so details are easier to see.
Histogram Equalization improves the contrast of an image by spreading out the brightness values. This makes dark areas lighter, bright areas clearer, and overall details more visible.
15. What is Feature Extraction? Name various techniques of Feature Extraction.
Feature Extraction means identifying and describing important parts of an image — such as edges, corners, shapes, colors, or textures — that help an AI system understand what is inside the image.
Different Feature Extraction techniques are:
- Edge Detection
- Corner Detection
- Texture Analysis
- Color Based Feature Extraction
- Deep Learning Based Feature Extraction
16. What is Edge Detection?
Edge detection is a technique in Computer Vision used to identify the boundaries or outlines of objects within an image.
Edges are places in an image where there is a sudden change in brightness, color, or intensity — for example, the border between a cat and the background.
17. What is Corner Detection?
Corner Detection is a technique used to find points in an image where two edges meet — basically, corners, junctions, or interest points.
Corners are important because they don’t change easily even if the image moves, rotates, or changes brightness — that’s why they’re used to track or match objects across images.
18. What is Texture Analysis?
Texture analysis studies the repeating patterns on surfaces — like how smooth, rough, regular, or irregular they are — using mathematical and structural methods.
19. What is Color-Based Feature Extraction?
Color-Based Feature Extraction is a technique used to identify and describe objects in an image based on their color information.
It helps a computer “see” and recognize things by analyzing color patterns, shades, and intensities instead of shapes or edges.
20. What is Deep Learning Based Feature Extraction?
Deep Learning–based Feature Extraction means using deep neural networks (especially Convolutional Neural Networks – CNNs) to automatically learn and extract important features (like edges, shapes, colors, and textures) from images.
21. What do you mean by Detection and Segmentation in Computer Vision?
Detection: Identifying and locating objects in an image (e.g., detecting cars on a road).
Segmentation: Dividing an image into meaningful regions or objects (e.g., separating background from foreground).
Types:
22. What are different types of Detection and Segmentation tasks?
There are two types of tasks:
- Single Object Tasks
- Multiple Object Tasks
23. What do you know about Single Object Tasks? Name them.
Single Object Tasks are computer vision tasks where the system focuses on only one main object in an image — not multiple objects. The goal is to understand or identify one object at a time, such as detecting, classifying, or locating it. There are mainly two methods:
- Classification
- Classification Localization
24. What is Classification?
It is the process to determine the category to which an object belongs. Classification assigns labels to images based on learned patterns (e.g., cat vs. dog).
25. Explain different Classification algorithms.
i. K- Nearest Neighbours (KNN): KNN is a simple algorithm that can classify objects on basis of their distance from labelled data points.
ii. Support Vector Machine (SVM): It is a machine learning algorithm that tries to draw the best possible line (or boundary) to separate different groups of data.
iii. Convolutional Neural Network (CNN): CNN is a type of deep learning algorithm mainly used for analyzing images (like recognizing faces, objects, or handwriting).
iv. K-Means Clustering: K-Means clustering is an unsupervised learning method that groups images/features into clusters based on similarity (e.g., grouping similar looking faces).
v. Autoencoders: Autoencoders are neural networks used to compress data and then reconstruct it again — like teaching a model to copy its own input but in a smaller, smarter way.
26. What is Classification + Localization?
Classification + Localization is a computer vision task that helps an AI model to identify an object and its location in an image.
It means the system not only recognizes the object (like “this is a dog 🐶”) but also draws a box around it to show where the dog is in the picture.
27. Explain different Localization methods?
i. Regional Proposal Network (RPN): It is a part of Convolutional Neural Network (CNN) models (like Faster R-CNN) that helps the system to find the location of an object in an image.
ii. YOLO (You Only Look Once):YOLO is an AI model that can detect and recognize multiple objects in an image or video in one single pass — meaning it “looks” at the image only once.
iii. SSD (Single Shot Detector): SSD is a deep learning algorithm used for object detection — it can detect and classify multiple objects in an image in one single step (or shot).
28. What do you know about Multiple Object Tasks? Name them.
Multiple object detection means an AI looks at a picture and finds all the different things in it. It tells what each thing is (like a car, person, or dog) and draws a box around each one to show where it is in the image. There are two primary tasks:
- Object Detection
- Image Segmentation
29. Explain Object Detection with respect to Multiple Object tasks. Explain its algorithms.
It is the process to identify and localize multiple objects in an image. There are four important algorithms:
i. Region Based Convolutional Neural Network (RCNN): RCNN finds possible object regions and uses CNN to identify what’s in each region. It is accurate but slow.
ii. Region Based Fully Convolutional Neural Network (RFCN): RFCN is a faster version of RCNN that uses convolution layers on the whole image instead of processing each region separately..
iii. YOLO (You Only Look Once): It splits the image into a grid and predicts boxes and labels at once. Very fast and works well for real-time detection.
iv. Single Shot Multi-Box Detector (SSD):SSD predicts boxes and labels in one step using several feature maps, giving a good balance between speed and accuracy.
30. What is Image Segmentation? Why is it important
Image segmentation is the process of dividing an image into meaningful regions or objects to make analysis easier. It separates background from objects of interest.
31. Name different types of Image Segmentation?
- Semantic Segmentation
- Instance Segmentation
32. Explain Semantic Segmentation?
Semantic Segmentation means labeling every pixel in an image by its type (like cat, dog, road), but it does not tell which individual object it belongs to.
Example:
Image: 3 cats in a photo
Semantic Segmentation Output:
All cat pixels are labeled “cat”, background is labeled “background”
The three cats are not separated — they all belong to the same class.
Use Cases:
Road lane detection
Medical imaging (e.g., tumor vs healthy tissue)
33. Explain Instance Segmentation?
Instance Segmentation is the process of classifying each pixel and distinguishing between different objects of the same category.
It tells the computer what type of object each pixel belongs to and which exact object it is.
Example:
Image: 3 cats in a photo
Instance Segmentation Output:
- Cat 1 → labeled “cat 1”
- Cat 2 → labeled “cat 2”
- Cat 3 → labeled “cat 3”
Each cat is individually recognized, even though they belong to the same class.
Use Cases:
- Self-driving cars (detecting multiple pedestrians separately)
- Counting objects in images
34. What is OpenCV?
OpenCV is a free library used in Python and other languages for image processing, computer vision, and tasks like face recognition and object detection.
35. How can we convert an image to Grayscale?
In OpenCV, we use: cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
This changes a colored (RGB) image into grayscale, where each pixel has only brightness values (0–255).
36. What is the main difference between Classification and Detection?
Ans:
- Classification: Identifies what is present in an image (e.g., “This image is of a dog”).
- Detection: Identifies what is present and where it is located by drawing bounding boxes (e.g., “Dog at top-left corner”).
37. Write down any two algorithms which can be used for Object Detection.
Ans:
- YOLO (You Only Look Once)
- SSD (Single Shot Detector)
38. Write down the process of Object Detection in a Single Object.
Ans:
- Image Acquisition: Capture the image.
- Preprocessing: Remove noise, adjust brightness/contrast.
- Feature Extraction: Identify edges, shapes, or patterns.
- Classification: Recognize if the object is present.
- Localization: Draw a bounding box around the detected object.
39. How does Noise Reduction enhance the quality of images in preprocessing?
Noise reduction removes unwanted parts or blurs from an image, making edges clearer, contrast better, and object detection more accurate.
40. Differentiate between Single Object Tasks and Multiple Object Tasks.
| Single Object Tasks | Multiple Object Tasks |
|---|---|
| Detects/classifies only one object in an image. | Detects/classifies many objects in the same image. |
| Example: Identifying if an image has a cat. | Example: Detecting people, cars, and traffic lights together in one image. |
41. What are some supervised and unsupervised learning algorithms used for classification?
Machine Learning algorithms used for classification include supervised learning and unsupervised learning. Some supervised learning algorithms are as follows:
- K-Nearest Neighbors (KNN): A non-parametric method that classifies objects based on their proximity to labelled data points.
- Support Vector Machines (SVM): Utilizes hyperplanes to separate objects into distinct classes.
- Convolutional Neural Networks (CNNs): Deep Learning models designed for high-accuracy image classification.
Some unsupervised learning algorithms are as follows:
- K-Means Clustering: Groups similar data points without prior labeling by minimizing intra-cluster variance.
- Autoencoders: A Deep Learning-based approach that learns representations of images for clustering.
42. What is High Level Processing in Computer Vision? What are its components
High level processing involves interpreting and understanding image content after preprocessing, feature extraction, and segmentation. Different components are
- Object Recognition: Identifying objects (e.g., detecting a cat or dog in an image).
- Scene Understanding: Understanding relationships among objects (e.g., traffic scene with cars, pedestrians, and signals).
- Context Analysis: Interpreting meaning by considering surrounding context (e.g., identifying whether a person is in a hospital or a classroom).
43. What are different applications of High-level Processing?
Autonomous Driving: Detecting lanes, vehicles, pedestrians.
Medical Diagnosis: Identifying tumors, diseases from scans.
Facial Recognition & Biometrics: Recognizing individuals for security, attendance, and authentication.
44. Explain applications of Computer Vision with examples.
- Facial Recognition: Used in phones, CCTV, and airports for identifying individuals.
- Healthcare and Medical Imaging: Detecting diseases in Xrays, MRIs, and CT scans.
- Autonomous Vehicles & ADAS: Self-driving cars detect roads, traffic signals, and obstacles.
- Optical Character Recognition (OCR): Converting printed/handwritten text into digital text (e.g., Google Lens, document scanning).
- Machine Inspection & Quality Control: Detecting defects in manufacturing (e.g., faulty products on assembly lines).
- 3D Model Construction & Reconstruction: Creating 3D models from 2D images (e.g., architecture, gaming, VR).
- Surveillance & Security: Detecting suspicious activities or intruders using CCTV + AI.
- Biometric Authentication & Identity Verification: Fingerprint, iris, and facial scans for secure access.
- Retail & Ecommerce: Virtual tryons, automated checkout (e.g., Amazon Go).
- Agriculture & Precision Farming: Monitoring crop health, detecting pests, automating harvesting.
45. What are the challenges of Computer Vision?
- Variation in Images: Lighting, angles, occlusion (objects blocking view).
- High Cost: Powerful hardware and large datasets.
- Data Quality : Blurry, noisy or biased images.
- Real Time Processing: Difficult to analyze highspeed videos in real time.
- Privacy & Ethical Concerns: Identity theft or biased recognition.
46. What is the future of Computer Vision?
- Integration with AI & IoT: Smart cities, smart homes with realtime vision systems.
- Improved Healthcare: Faster and more accurate disease detection.
- Enhanced Autonomous Vehicles: Safer navigation and traffic management.
- Virtual Reality: Realistic interactions and immersive experiences.
- Agriculture: AI powered drones for better crop monitoring.
- Retail : AI powered virtual assistants and automated stores.
- Defense: Smarter surveillance and threat detection.
47. A traffic monitoring system is being developed to detect vehicles in real time. However, the images captured by the roadside cameras often contain noise due to poor lighting conditions at night. What preprocessing techniques can be used to improve the image quality before vehicle detection, and why?
- Noise Reduction: Removes unwanted distortions (grains, speckles).
- Contrast Adjustment: Makes vehicles more visible.
- Histogram Equalization: Improves brightness and contrast.
- Deepfake Videos: Concerns & Role of Computer Vision
48. A news verification organization is addressing the issue of manipulated videos on social media. These videos, often created using advanced Computer Vision techniques, can make public figures appear to say or do things they never did. What are the primary concerns associated with such deepfake content, and how does Computer Vision both enable and help detect these manipulations?
Concerns:
- Misinformation: Fake news and false events.
- Loss of Trust: People may stop believing real videos.
- Social/Political Impact: Can harm reputations and stability.
Role of Computer Vision:
- Enables Deepfakes: Uses AI to create fake but realistic videos.
- Detects Deepfakes: Finds mismatches in lighting, facial expressions, or movements to identify fake content.
49. A social media platform uses facial recognition to automatically tag users in photos. However, privacy advocates raise concerns about its ethical implications. What are some challenges associated with facial recognition in public spaces, and how can they be addressed?
Challenges:
- Privacy Issues: People may be tracked without consent.
- Bias in AI Models: Misidentification due to unbalanced training data.
- Security Risks: Risk of data misuse or hacking.
Solutions:
- Strong data protection laws.
- Transparent rules for AI use.
- Use diverse datasets to reduce bias.
50. An online art exhibition is being organized where artists must upload high-quality digital images of their paintings. Riya, an artist, is worried that her uploaded painting might look blurry when viewers zoom in. Explain to Riya how bit depth affects the color quality of her painting and why it is important for digital viewing.
Answer:
- Bit depth decides how many colors or shades can be represented in an image.
- A higher bit depth means smoother color transitions and more accurate details.
- Low bit depth can show visible lines in gradients.
- High bit depth ensures realistic colors and finer details.
51. At an international airport, an advanced AI-based monitoring system is installed to ensure passenger safety. Different feature extraction methods are applied to detect unusual activities. Identify the suitable feature extraction techniques in the following situations:
i. A suitcase is detected left behind near the check-in counter.
ii. A person suddenly runs across a restricted area.
iii. Security staff are asked to identify a missing child wearing a bright yellow jacket.
Answer:
i. Object Detection / Segmentation → To track unattended luggage.
ii. Motion Detection / Edge Detection → To recognize sudden unusual movement.
iii. Color-based Feature Extraction → To identify the child by the yellow jacket.
52. Drones use Computer Vision to deliver packages and monitor large areas. What problems could occur if the drone misidentifies objects (e.g., confusing a tree with a landing spot)? What methods can be applied to prevent such errors?
Problems if errors occur:
- Drone may crash into trees or buildings.
- Package may fall at the wrong place.
- People or property may get harmed.
Methods to prevent errors:
- Use more than one sensor (camera + GPS + LiDAR).
- Train with many images in different conditions.
- Add safety checks to detect mistakes.
- If unsure, the drone should stop or return to base.
53: A self driving car uses Computer Vision to recognize road signs, pedestrians, and other vehicles. How might Computer Vision errors (such as misinterpreting a stop sign) affect real-life situations? What safety measures could developers implement to reduce such risks?
Problems if errors happen:
- Misreading a stop sign may cause accidents.
- Not seeing a pedestrian can lead to injury.
- Wrongly judging another vehicle can cause collisions.
Safety measures:
- Use multiple sensors (cameras, LiDAR, radar) to check.
- Train the system with many different images (day, night, rain, damaged signs).
- Monitor the system in real time and alert drivers.
- If unsure, the car should slow down or stop automatically.