Image Processing Basics: Fundamentals, Types, and Applications

1. What is an Image?

An image is a 2D function $f(x,y)$ where $x$ and $y$ are spatial coordinates, and the amplitude $f$ represents the intensity (e.g., brightness or color) at each point (pixel). Images can be classified based on their dimensionality and color representation:

1.1 Types of Images

classDiagram
    class Image {
        <<abstract>>
        +dimensions: int
        +colorModel: string
    }
    class GrayscaleImage {
        +intensity: 0-255
    }
    class ColorImage {
        +model: RGB/CMYK/HSV
    }
    class BinaryImage {
        +threshold: 0 or 1
    }

• Grayscale Image: Single-channel (0–255), e.g., X-ray scans.
• Color Image: Multi-channel (RGB, CMYK), e.g., photos from smartphones.
• Binary Image: Black-and-white (0 or 1), e.g., OCR (Optical Character Recognition) outputs.

2. Digital Image Formation

• Resolution: Number of pixels (e.g., 1920×1080 for Full HD).
• Bit Depth: Bits per pixel (e.g., 8-bit = 256 grayscale levels, 24-bit = 16.7M colors).
• Dynamic Range: Ratio of brightest to darkest detectable intensity.

Worked Example: Pixel Calculation A 4K image has dimensions 3840×2160. Calculate:

1. Total pixels = $3840\times 2160=8,294,400$ pixels.
2. File size (8-bit grayscale) = $8,294,400\times 1$ byte = 8.29 MB.
3. File size (24-bit RGB) = $8,294,400\times 3$ bytes = 24.88 MB.

3. Image Processing Systems

An image processing system consists of:

1. Input: Sensor (camera, scanner) or synthetic data (3D rendering).
2. Processing: Algorithms (enhancement, segmentation, recognition).
3. Output: Display, storage, or further analysis.

flowchart LR
    A["Input: Camera/Scanner"] --> B["Preprocessing: Noise Reduction"]
    B --> C["Enhancement: Contrast Adjustment"]
    C --> D["Segmentation: Object Extraction"]
    D --> E["Recognition: Classification"]
    E --> F["Output: Display/Storage"]

In the Real World

• eSewa: Uses image processing to verify fingerprint scans (binary segmentation) for user authentication.
• Pathao: Applies edge detection (Canny filter) to analyze driver routes from satellite images for optimal delivery paths.
• NTC Traffic Cameras: Use motion detection (frame differencing) to identify traffic violations in real time.

4. Applications of Image Processing

5. Workflow of Image Processing

A typical pipeline:

1. Acquisition: Capture image (e.g., via webcam or medical scanner).
2. Preprocessing: Correct distortions (e.g., noise removal).
3. Enhancement: Improve visual quality (e.g., histogram equalization).
4. Segmentation: Partition image into regions (e.g., foreground/background).
5. Description/Recognition: Extract features (e.g., SIFT for object matching).

Worked Example: Traffic Sign Recognition (Nepal)

1. Input: Blurry traffic sign photo from a dashboard camera.
2. Preprocessing: Apply Gaussian blur to reduce noise.
3. Enhancement: Sharpen using unsharp masking.
4. Segmentation: Use Otsu’s thresholding to binarize the sign.
5. Recognition: Train a CNN to classify signs (e.g., "Stop" vs. "Speed Limit").

graph LR
    A["Input: Blurry Sign"] --> B["Gaussian Blur"]
    B --> C["Unsharp Masking"]
    C --> D["Otsu Thresholding"]
    D --> E["CNN Classification"]
    E --> F["Output: 'Stop Sign'"]

6. Challenges in Image Processing

Exam Tip

1. Definitions: Know the difference between grayscale, color, and binary images (e.g., "Binary images use a threshold to separate foreground/background").
2. Calculations: Practice pixel count and file size problems (e.g., "A 1024×768 RGB image has 24-bit depth. Calculate its size in MB").
3. Applications: Link techniques to real-world examples:
   • eSewa: Binary segmentation for fingerprint verification.
   • Pathao: Edge detection for route optimization.
4. Diagrams: Draw the image processing pipeline (input → preprocessing → enhancement → segmentation → recognition → output) in exams.
5. Common Pitfalls:
   • Confusing resolution (pixels) with bit depth (color levels).
   • Forgetting that color images require 3 channels (RGB), not 1.

Key Formulae to Memorize

1. Total pixels = $\text{width}\times \text{height}$.
2. File size (bytes) = $\text{pixels}\times \text{bit depth}/8$.
3. Dynamic range = ${\log }_2(\text{bit depth})$.