Image Recognition

📌 What Is Image Recognition?

Image recognition is the process of teaching a computer to:

• Look at an image
• Understand what is inside it
• Classify or identify objects

Example:

• Input → Picture of a cat
• Output → “Cat”

Modern image recognition is powered by deep learning, especially Convolutional Neural Networks (CNNs).

How Does Image Recognition Work?

At a high level:

1. Input Image
2. Feature Extraction (edges, shapes, patterns)
3. Classification
4. Output Prediction

Step 1: Image as Numbers

A computer does not see images like humans.

It sees:

• A grid of pixels
• Each pixel has numbers

Example:

• Grayscale image → values from 0 to 255
• RGB image → 3 channels (Red, Green, Blue)

Example shape:

Image shape = (Height, Width, Channels)
Example = (224, 224, 3)

Step 2: Feature Extraction (CNN)

This is where Convolutional Neural Networks (CNNs) come in.

CNNs:

• Detect edges
• Detect textures
• Detect shapes
• Combine features into objects

Convolution Layer

A small filter (kernel) slides over the image.

Example 3×3 filter:

[ 1  0 -1
  1  0 -1
  1  0 -1 ]

This filter detects vertical edges.

The filter moves across the image and creates a feature map.

Pooling Layer

Reduces size of feature map.

Example:

• MaxPooling(2×2)
• Takes maximum value in each 2×2 region

Purpose:

• Reduce computation
• Prevent overfitting
• Keep important features

Typical CNN Architecture

Input Image
↓
Convolution
↓
ReLU
↓
Pooling
↓
Fully Connected Layer
↓
Softmax (Output)

Popular CNN Architectures

Below is a markdown-ready table:

| Architecture | Year | Key Idea | Created By |
|--------------|------|----------|------------|
| LeNet-5 | 1998 | First CNN for digits | Yann LeCun |
| AlexNet | 2012 | Deep CNN breakthrough | Alex Krizhevsky |
| VGG16 | 2014 | Very deep simple layers | Oxford |
| ResNet | 2015 | Skip connections | Microsoft |
| EfficientNet | 2019 | Scaled CNN design | Google |

🔥 Breakthrough Moment

The big breakthrough came in 2012 when ImageNet competition was won by AlexNet.

This showed deep learning dramatically outperformed traditional computer vision.

Modern Image Recognition Systems

Used in:

• Self-driving cars
• Medical imaging
• Face recognition
• Security systems

Major AI research companies involved:

• Google
• OpenAI
• Microsoft

Example: Simple CNN in PyTorch

import torch
import torch.nn as nn
import torch.nn.functional as F

class ImageClassifier(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(3, 16, 3)  # RGB input
        self.pool = nn.MaxPool2d(2, 2)
        self.fc1 = nn.Linear(16 * 111 * 111, 10)

    def forward(self, x):
        x = self.pool(F.relu(self.conv1(x)))
        x = x.view(x.size(0), -1)
        x = self.fc1(x)
        return x

model = ImageClassifier()
print(model)

This model:

• Takes 3-channel image
• Applies convolution
• Pools features
• Classifies into 10 categories

Training Process

1. Load dataset (images + labels)
2. Forward pass
3. Compute loss
4. Backpropagation
5. Update weights
6. Repeat for many epochs

Loss Function for Image Recognition

For classification:

CrossEntropyLoss

This combines:

• Softmax
• Log loss

How Large-Scale Systems Work

Large models:

• Use millions of images
• Use GPUs
• Use data augmentation
• Use pretrained models

Example pretrained models:

• ResNet
• EfficientNet
• Vision Transformer

CNN vs Vision Transformer

| CNN | Vision Transformer |
|------|--------------------|
| Uses convolution filters | Uses self-attention |
| Strong local feature extraction | Better global understanding |
| Data-efficient | Needs large datasets |

Real-World Applications

• Face Unlock (smartphones)
• Cancer detection from X-rays
• Traffic sign detection
• Product recommendation via images

Full Combined Code Example

Below is a minimal training-ready example: