Linear Filtering
1. Introduction
- Linear filtering processes an image by convolving it with a kernel (filter mask) to enhance
or extract
features.
- It helps in noise reduction, edge detection, and image sharpening.
- The output pixel value is computed as a weighted sum of the pixel's neighborhood defined
by the filter
kernel.
2. Types of Linear Filters
- Low Pass Filters: Used for smoothing and noise reduction by averaging pixel values.
- High Pass Filters: Used for edge detection by highlighting intensity changes in the image.
A. Low Pass Filters (Smoothing Filters)
1. Mean Filter
- Replaces each pixel with the average of its surrounding pixels, reducing noise.
- Results in image blurring due to smoothing of details.
Mean Filter Kernel (3x3):
[111
111
1 1 1 ] x 1/9
Example Image Patch:
[ 10 20 30
40 50 60
70 80 90 ]
Sum = 450
Average = 450 / 9 = 50
Advantages:
- Simple and easy to implement.
- Effectively reduces random noise.
Disadvantages:
- Blurs sharp edges and fine details.
- Not good for images requiring edge preservation.
2. Gaussian Filter
- Uses weighted average giving more importance to central pixels, preserving edges better.
- More effective in smoothing while maintaining image structure.
Gaussian Filter Kernel (3x3):
[121
242
1 2 1 ] x 1/16
Example Image Patch:
[ 100 100 100
100 200 100
100 100 100 ]
Weighted sum = 2000
Output = 2000 / 16 = 125
Advantages:
- Better at preserving edges than mean filter.
- Reduces noise while keeping image features intact.
Disadvantages:
- More complex computation than mean filter.
- Slight blurring still occurs.
B. High Pass Filters (Edge Detection)
1. Laplacian Filter
- Detects edges by calculating the second derivative of the image.
- Highlights regions with rapid intensity changes in all directions.
Laplacian Kernel:
[ 0 -1 0
-1 4 -1
0 -1 0 ]
Example:
[ 0 0 0
0 255 0
0 0 0]
Output = 4 × 255 = 1020 (clipped to 255 in 8-bit images)
Advantages:
- Detects edges in all directions.
- Simple and effective for edge detection.
Disadvantages:
- Sensitive to noise.
- Does not provide edge direction information.
2. Sobel Filters
- Approximates the first derivative to detect horizontal and vertical edges.
- Combines smoothing and differentiation, reducing noise sensitivity.
Sobel X (Vertical Edges):
[ -1 0 1
-2 0 2
-1 0 1 ]
Sobel X Example:
[ 10 10 10
20 20 20
30 30 30 ]
Output = 0 (No vertical edge)
Sobel Y (Horizontal Edges):
[ -1 -2 -1
0 0 0
1 2 1]
Sobel Y Example:
[ 10 10 10
20 20 20
30 30 30 ]
Output = 80 (Horizontal edge)
Advantages:
- Less sensitive to noise due to smoothing.
- Provides edge magnitude and direction.
Disadvantages:
- More computationally intensive than Prewitt.
- May miss fine edges.
3. Prewitt Filters
- Similar to Sobel with simpler weights.
- Detects vertical and horizontal edges using uniform difference.
Prewitt X (Vertical Edges):
[ -1 0 1
-1 0 1
-1 0 1 ]
Prewitt X Example:
[ 10 10 10
20 20 20
30 30 30 ]
Output = 0 (No vertical edge)
Prewitt Y (Horizontal Edges):
[ -1 -1 -1
0 0 0
1 1 1]
Prewitt Y Example:
[ 10 10 10
20 20 20
30 30 30 ]
Output = 60 (Horizontal edge)
Advantages:
- Simpler and faster than Sobel.
- Effective for edge detection.
Disadvantages:
- More sensitive to noise than Sobel.
- Less accurate edge direction information.
