Chapter 10: Computer Vision (CV)

"Teaching machines to see and interpret images/videos like humans."

---

🔹 1. What is Computer Vision?

Computer Vision (CV) is a subfield of AI that allows machines to analyze, process, and understand visual data (images, videos).

📷 Applications:

• Face recognition (e.g., iPhone Face ID)

• Object detection (e.g., self-driving cars)

• Medical imaging (e.g., tumor detection)

• OCR (Optical Character Recognition)

---

🔹 2. Basic Image Concepts

Term	Meaning
Pixel	Smallest unit of an image
Grayscale Image	Single channel (Black & White)
RGB Image	3 channels (Red, Green, Blue)
Resolution	Width × Height of an image
Image Matrix	Each pixel represented as a number (0-255)

📌 Images are represented as numerical arrays → can be processed with ML/DL.

---

🔹 3. Image Processing with OpenCV

OpenCV is a popular open-source library for computer vision tasks.

Common Operations:

python
import cv2
img = cv2.imread('dog.jpg')          # Load image
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)  # Convert to grayscale
blurred = cv2.GaussianBlur(gray, (5,5), 0)    # Apply blur
edges = cv2.Canny(blurred, 100, 200)          # Edge detection

---

🔹 4. Object Detection vs. Image Classification

Task	Goal	Example
Classification	Identify object	"This is a cat"
Detection	Locate object + identify	"Cat is at (x1,y1,x2,y2)"
Segmentation	Pixel-level classification	Separate each pixel of cat vs background

---

🔹 5. Deep Learning for CV

Most modern CV tasks are done using Convolutional Neural Networks (CNNs).

🔸 CNN (Convolutional Neural Network)

• Detect patterns (edges, corners, shapes)

• Layers:

  • Convolution Layer → Feature extractor

  • Pooling Layer → Downsampling

  • Fully Connected Layer → Classification

Example Architecture:

mathematica
Input Image → Conv → ReLU → Pool → Conv → ReLU → Pool → FC → Output

Libraries:

• TensorFlow/Keras

• PyTorch

---

🔹 6. Image Classification Example with Keras

python
from tensorflow.keras.datasets import mnist
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense

(X_train, y_train), (_, _) = mnist.load_data()
X_train = X_train.reshape(-1, 28, 28, 1) / 255.0

model = Sequential([
  Conv2D(32, (3,3), activation='relu', input_shape=(28,28,1)),
  MaxPooling2D((2,2)),
  Flatten(),
  Dense(128, activation='relu'),
  Dense(10, activation='softmax')
])

model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
model.fit(X_train, y_train, epochs=5)

---

🔹 7. Advanced CV Techniques

🔸 Object Detection

Model	Use Case
YOLO (You Only Look Once)	Real-time detection
SSD (Single Shot Detector)	Faster, less accurate
Faster R-CNN	High accuracy, slower

🔸 Image Segmentation

Technique	Tool
Semantic Segmentation	Classify each pixel
Instance Segmentation	Separate object instances
Mask R-CNN	Does both!

🔸 Face Detection & Recognition

• Haar Cascades (OpenCV)

• FaceNet, Dlib, DeepFace (deep learning)

---

🔹 8. Projects in Computer Vision

Project	Tools Used
Face Mask Detector	CNN + OpenCV
Number Plate Recognition	OCR + Tesseract
Emotion Recognition	CNN + Facial landmarks
Real-Time Object Detection	YOLO + Webcam
AI Virtual Mouse	Hand tracking with MediaPipe

---

🔹 9. Tools for Practice

Tool	Purpose
OpenCV	Image/video processing
MediaPipe	Real-time hand/face tracking
LabelImg	Annotating datasets
Kaggle	Practice datasets
Google Colab	Free GPU for training

---

✅ Chapter Summary

Topic	Key Learning
CV Basics	Pixels, channels, image formats
OpenCV	Preprocessing, transformations
CNN	Most powerful for image tasks
Detection Models	YOLO, SSD, Mask R-CNN
Real-world Use	Face recognition, object detection

---

💡 Mini Tasks:

1. Build an MNIST digit classifier using CNN.

2. Create a real-time face detector using OpenCV.

3. Train a YOLOv5 model on custom images (with LabelImg).

4. Try image segmentation using Mask R-CNN or DeepLab.