Overview of Computer Vision: Concept and Applications — Beginner-Friendly Notes
1. What Computer Vision (CV) Is
CV = AI that sees — extracts meaning from images/video (pixels → objects, scenes, actions). It is the AI "eyes".
2. Pipeline (concept)
Image (pixels) → Preprocessing → Feature Extraction (CNN) → Task head → Output (labels/boxes/masks)
Modern core is CNN (Convolutional Neural Network) and now Vision Transformers, learned from labelled image data.
3. Core Tasks (know the difference)
| Task | Output | Example |
|---|---|---|
| Classification | one label per image | cat vs dog |
| Object Detection | boxes + labels | "pedestrian at (x,y)" for self-driving |
| Segmentation | pixel-wise mask | tumour boundary in MRI |
| Face recognition | identity | phone unlock, attendance |
| Pose / Action | skeleton, action label | "person waving" |
| OCR | text from image | reading number plate, handwritten notes |
4. Why Hard
- Variation: lighting, angle, occlusion, scale.
- Needs large labelled data + robustness.
- Must handle real-time (30 fps) for driving/robotics.
5. Applications — Exam List (branch-relevant)
- Autonomous: lane/pedestrian detection, driver monitoring.
- Healthcare: X-ray/retina diagnosis (TB, diabetic retinopathy at PHCs).
- Agriculture: crop disease, weed detection from drone images (India).
- Manufacturing: defect detection on line (mechanical).
- Civil/Surveillance: crack detection, crowd counting.
- Everyday: phone camera (portrait, scene), AR filters, search by image.
6. One-Line Summary
Computer vision uses CNNs to transform pixels into tasks like classification, detection and segmentation, enabling eyes for applications from self-driving and medical diagnosis to agriculture and manufacturing inspection.
7. Quick Self-Check
- Detection vs segmentation — what's the difference?
- Name one healthcare and one agriculture CV use in India.