COMPLETE IMAGE PROCESSING MASTER
NOTES
This PDF contains detailed theory, mathematical formulas, applications, advantages, disadvantages, types,
comparisons, numerical examples, and important concepts discussed in Unit-1 Fundamentals of Image Processing.
1. Introduction to Image Processing
Definition:
Image processing is the manipulation and analysis of digital images using mathematical operations and algorithms to
improve image quality or extract information.
Main Objectives: Enhancement Restoration Compression Segmentation Feature extraction Applications: Medical
imaging Satellite imaging Face recognition Autonomous vehicles Remote sensing Industrial inspection Advantages:
Improves image quality Helps automation Enhances analysis accuracy Disadvantages: Computationally expensive
Noise sensitive Requires large storage
2. What is an Image?
Definition:
An image is a two-dimensional representation of visual information composed of pixels.
Types of Images: Binary Image Grayscale Image Color Image Multispectral Image Digital Image Representation:
f(x,y) where x and y are spatial coordinates and f is intensity.
Image Properties: Resolution Intensity Brightness Contrast
3. Color Models
RGB Model: Uses Red, Green and Blue components.
Applications: Displays, monitors, cameras.
CMYK Model: Cyan, Magenta, Yellow and Black.
Applications: Printing.
HSV Model: Hue, Saturation, Value.
Applications: Computer vision and object tracking.
YCbCr Model: Used in video compression.
Advantages: Better color representation Supports compression and segmentation
4. Image Processing vs Computer Vision
Image Processing: Input and output are images Focuses on enhancement and transformations Computer Vision:
Extracts understanding from images Output is semantic interpretation Examples: Image Processing: Noise removal
Computer Vision: Face detection
5. Image File Formats
BMP: Uncompressed image format.
Advantage: High quality.
Disadvantage: Large size.
JPEG: Lossy compression.
Advantage: Small file size.
Disadvantage: Quality loss.
PNG: Lossless compression and transparency support.
GIF: Supports animation.
TIFF: High-quality professional image format.
6. Contrast Enhancement
Definition:
Contrast enhancement improves image visibility by increasing intensity differences.
Techniques: Contrast Stretching Histogram Equalization CLAHE Gamma Correction Contrast Stretching Formula:
s = ((r-rmin)/(rmax-rmin)) × (L-1)
Gamma Correction Formula:
s = c × r^gamma
Applications: Medical imaging Satellite imaging Photography Advantages: Better visibility Improved feature extraction
Disadvantages: Can amplify noise May over-enhance image
7. Histogram Equalization
Definition:
Histogram equalization redistributes intensity values to improve global contrast.
Formula:
s_k = (L-1) × SUM[p(r_j)]
Steps: Compute histogram Compute PDF Compute CDF Apply transformation Advantages: Automatic enhancement
Improves visibility Disadvantages: Can amplify noise Brightness may change Numerical Example:
Histogram:
0 → 4 pixels
1 → 3 pixels
2 → 1 pixel
Total pixels = 8
PDF for intensity 0 = 4/8 = 0.5
CDF = cumulative probabilities.
Transformation redistributes intensities uniformly.
8. Histogram Specification
Definition:
Histogram specification transforms image histogram to match a desired histogram.
Steps: Compute input histogram Compute target histogram Compute CDFs Match CDF values Map intensities
Applications: Medical imaging Standardized appearance correction Advantages: Better control than equalization
Custom enhancement Disadvantages: More complex Needs target histogram Numerical Example:
Input mapping:
0→2
1→2
2→3
3→3
9. Image Noise and Types
Definition:
Noise is unwanted variation in pixel intensity values.
Noise Models:
Additive: g(x,y)=f(x,y)+n(x,y)
Multiplicative: g(x,y)=f(x,y)×n(x,y)
Types of Noise: Gaussian Noise Salt-and-Pepper Noise Speckle Noise Poisson Noise Periodic Noise Quantization
Noise Gaussian Noise:
Follows Gaussian distribution.
Cause: Sensor electronics.
Salt-and-Pepper Noise:
Random black and white pixels.
Cause: Transmission errors.
Speckle Noise:
Multiplicative granular noise.
Applications: Radar and ultrasound.
Advantages of Noise Analysis: Improves denoising techniques Enhances image quality
10. Spatial Domain Filtering
Definition:
Spatial filtering operates directly on image pixels using kernels/masks.
Types: Smoothing Filters Sharpening Filters Mean Filter:
Replaces center pixel with neighborhood average.
Formula: Sum of neighborhood / number of pixels.
Gaussian Filter:
Weighted smoothing filter.
Formula: G(x,y)=1/(2*pi*sigma^2)e^(-(x^2+y^2)/(2*sigma^2))
Median Filter:
Replaces center pixel with median value.
Best for salt-and-pepper noise.
Laplacian Filter:
Second derivative sharpening filter.
Formula: ∇²f = d²f/dx² + d²f/dy²
Sobel Filter:
Gradient-based edge detector.
Magnitude: |G| = sqrt(Gx² + Gy²)
Advantages: Effective denoising Edge enhancement Disadvantages: Can blur details Sharpening amplifies noise
11. Numerical Examples on Filters
Mean Filter Example:
Neighborhood:
10 20 30
20 40 60
30 60 90
Sum = 360
Mean = 360/9 = 40
Median Filter Example:
10 20 30
20 255 40
30 40 50
Sorted values:
10 20 20 30 30 40 40 50 255
Median = 30
Laplacian Example:
Using mask:
0 -1 0
-1 4 -1
0 -1 0
Filtered output = 160
Sobel Example:
Strong gradients indicate edges.
Edge magnitude calculated using sqrt(Gx² + Gy²).
12. Frequency Domain Filtering
Definition:
Frequency domain filtering processes image frequency components using Fourier Transform.
Steps: Apply Fourier Transform Generate frequency spectrum Apply filter Apply inverse transform DFT Formula:
F(u,v)=SUMSUM[f(x,y)e^(-j2pi(ux/M+vy/N))]
Types of Filters: Low-pass filter High-pass filter Band-pass filter Band-reject filter Notch filter Low-pass Filter:
Removes high frequencies and smooths image.
High-pass Filter:
Enhances edges and details.
Gaussian LPF:
H(u,v)=e^(-D²(u,v)/(2D0²))
Advantages: Efficient for large kernels Excellent periodic noise removal Disadvantages: Mathematically complex Can
produce ringing artifacts
13. Important Formula Sheet
1. g(x,y)=f(x,y)+n(x,y)
2. p(r_k)=n_k/n
3. CDF(r_k)=SUM[p(r_j)]
4. s_k=(L-1)SUM[p(r_j)]
5. PSNR=10log10(MAX²/MSE)
6. MSE=(1/MN)SUM[(f-g)²]
7. |G|=sqrt(Gx²+Gy²)
8. H(u,v)=e^(-D²(u,v)/(2D0²))
9. ∇²f = d²f/dx² + d²f/dy²
FINAL REVISION POINTS
Histogram equalization improves global contrast. Median filter is best for salt-and-pepper noise. Gaussian filter is best
for Gaussian noise. Laplacian filter sharpens images. Sobel and Prewitt are edge detectors. Low-pass filters smooth
images. High-pass filters enhance details. Frequency domain uses Fourier Transform.