Image Enhancement and Filtering

CMSC 178IP - Module 04

Noel Jeffrey Pinton
Department of Computer Science
University of the Philippines Cebu

Image Enhancement and Filtering

CMSC 178IP - Module 04

Noel Jeffrey Pinton
Department of Computer Science
University of the Philippines Cebu

Learning Objectives

By the end of this module, you will be able to:

Identify different types of image noise
Apply spatial domain smoothing filters
Implement frequency domain filtering
Use edge-preserving filters for denoising
Apply sharpening techniques to enhance edges

Noise in Images

Understanding image degradation

Types of Image Noise

Gaussian Noise
Additive, bell-shaped distribution

Salt & Pepper
Random black/white pixels

Noise Characteristics

Additive Noise

$$g(x,y) = f(x,y) + n(x,y)$$

Gaussian noise
Uniform noise

Multiplicative Noise

g(x,y) = f(x,y) \cdot n(x,y)

Speckle noise
Film grain

Knowledge Check

Think About It

What type of noise is best removed by a median filter?

Click the blurred area to reveal the answer

Smoothing Filters

Spatial domain noise reduction

Smoothing Filters Comparison

Comparison of mean, Gaussian, and median filters

Low-Pass Filters

Box Filter (Mean)

1/9 * [[1,1,1],
       [1,1,1],
       [1,1,1]]

Gaussian Filter

1/16 * [[1,2,1],
        [2,4,2],
        [1,2,1]]

Denoising Results

Different filters have different strengths for various noise types

Edge-Preserving Filters

Smoothing without blurring edges

Bilateral Filter

Bilateral Filter: Combines spatial proximity AND intensity similarity weights.

w(i,j,k,l) = \exp\left(-\frac{(i-k)^2+(j-l)^2}{2\sigma_s^2}\right) \cdot \exp\left(-\frac{|I(i,j)-I(k,l)|^2}{2\sigma_r^2}\right)

Knowledge Check

Think About It

Why does the bilateral filter preserve edges while smoothing?

Click the blurred area to reveal the answer

Adaptive Filtering

Adaptive filters adjust their behavior based on local image statistics (variance, mean).

Sharpening Filters

Enhancing edges and fine details

High-Pass Filters

Laplacian

[[ 0,-1, 0],
 [-1, 4,-1],
 [ 0,-1, 0]]

Sobel (X direction)

[[-1, 0, 1],
 [-2, 0, 2],
 [-1, 0, 1]]

Sharpening Techniques

Unsharp Masking:

g = f + k \cdot (f - f_{blurred})

where k controls sharpening strength

Edge Detection

Comparison of Sobel, Prewitt, and Canny edge detectors

Knowledge Check

Think About It

What makes the Canny edge detector superior to simple gradient methods?

Click the blurred area to reveal the answer

Filter Implementation

import cv2
import numpy as np

img = cv2.imread('image.jpg', 0)

# Gaussian blur
blurred = cv2.GaussianBlur(img, (5,5), 1.5)

# Median filter (good for salt & pepper)
median = cv2.medianBlur(img, 5)

# Bilateral filter (edge-preserving)
bilateral = cv2.bilateralFilter(img, 9, 75, 75)

# Sharpening with Laplacian
laplacian = cv2.Laplacian(img, cv2.CV_64F)
sharpened = img - 0.5 * laplacian

Summary

Key Takeaways

Noise types: Gaussian (additive), salt & pepper (impulse), speckle (multiplicative)
Smoothing filters: Mean, Gaussian, median - trade-off between noise reduction and blur
Low-pass: Removes high frequencies (noise, edges)
Bilateral filter: Edge-preserving smoothing using intensity similarity
High-pass/sharpening: Enhances edges using Laplacian or unsharp masking
Edge detection: Canny provides best results with non-maximum suppression

Questions?

Thank you for your attention!

Next: Module 05 - Image Restoration

End of Module 04

Image Enhancement and Filtering

Questions?