Image Filtering

Definition of Image Filtering

• Image filtering is a technique used to modify or enhance
an image by applying a filter, which is typically
represented by a kernel or a mask. Filters can smooth
images, highlight features like edges, or reduce noise.
Noise
• Noise in image processing refers to random variations in
pixel values caused by errors in image acquisition,
transmission, or storage.
• Noise distorts image quality and makes image analysis
and processing tasks more challenging.
Sources of Noise
Noise is caused by any of the following:
1. Sensor Limitations: Imperfections in imaging
hardware, such as CCD or CMOS sensors.
2. Environmental Factors: Variations in lighting,
temperature, or electrical interference.
3. Transmission Errors: Signal degradation or data loss
during transmission.
4. Quantization Noise: Occurs when analog signals are
converted to digital form.
5. Compression Artifacts: Errors introduced by lossy
image compression methods.
Noise Models
• Mathemtically, noise can be represented as:

• Where
• Noisy image
• Original image
• Noise component
• For multiplicative noise (e.g., speckle noise):
• .N
Types of Noise
• Noise is grouped by the following:
1. Gaussian Noise
2. Salt-and-pepper Noise (Impulse noise)
3. Speckle Noise
4. Poisson Noise (Short Noise)
5. Periodic Noise
Gaussian Noise
• Gaussian noise, also known as normal noise, follows a
normal distribution. It is characterized by its mean (μ)
and standard deviation (σ), which determine its
intensity and spread. Mathematically, Gaussian noise is
represented as:
Causes of Gaussian Noise
• Thermal Noise: Generated by sensors due to
temperature.
• Electronic Noise: Occurs due to electronic components
in imaging devices.
• Acquisition Noise: Results from imperfections in the
image acquisition process.
Characteristics of Gaussian Noise
Characteristics of Gaussian Noise
1.Normal Distribution: Follows the bell-curve
probability distribution.
2.Additive Nature: Can be modeled as:

Inoisy = Ioriginal+N

3. Wide Applicability: Commonly used in image

processing for testing and simulating real-world scenarios
Effects of Gaussian Noise
• Distortion: Blurs edges and fine details.
• Performance Degradation: Reduces the efficiency of
segmentation, edge detection, and feature extraction.
Reducing Gaussian Noise
1.Spatial Domain Filters:
1.Gaussian Blur: Smoothens the image while preserving
edges.
2.Bilateral Filter: Combines spatial and intensity information
to preserve edges better.
2.Frequency Domain Filters:
1.Low-Pass Filters: Suppress high-frequency components
associated with noise.
3.Advanced Techniques:
1.Non-Local Means, Wavelet Transform, and Deep Learning
models
Practical Implementation
12. # Display original and noisy images
• (a) Adding Gaussian Noise
13. plt.figure(figsize=(10, 5))
1. import cv2
2. import numpy as np 14. plt.subplot(1, 2, 1)
3. from matplotlib import pyplot as plt
4. # Load the image 15. plt.imshow(image, cmap='gray’)
5. image = cv2.imread('example.jpg', 16. plt.title('Original Image’)
cv2.IMREAD_GRAYSCALE)
6. # Add Gaussian noise 17. plt.axis('off’)
7. mean = 0
8. std_dev = 20 # Standard deviation 18. plt.subplot(1, 2, 2)
9. gaussian_noise = np.random.normal(mean,
std_dev, image.shape).astype(np.float32) 19. plt.imshow(noisy_image, cmap='gray’)
10. noisy_image =
cv2.add(image.astype(np.float32), 20. plt.title('Gaussian Noisy Image’)
gaussian_noise)
21. plt.axis('off’)
11. noisy_image =
np.clip(noisy_image, 0, 255).astype(np.uint8)
22. plt.show()
Removal of Gaussian Noise with
filters
Using Gaussian Blur
1. # Apply Gaussian blur
2. blurred_image = cv2.GaussianBlur(noisy_image, (5, 5), 0)

3. # Display the results

4. plt.figure(figsize=(10, 5))
5. plt.subplot(1, 2, 1)
6. plt.imshow(noisy_image, cmap='gray')
7. plt.title('Noisy Image')
8. plt.axis('off')

9. plt.subplot(1, 2, 2)
10. plt.imshow(blurred_image, cmap='gray')
11. plt.title('Gaussian Blurred Image')
12. plt.axis('off')

13. plt.show()
1. # Apply bilateral filter
2. bilateral_filtered = cv2.bilateralFilter(noisy_image, 9, 75, 75)

3. # Display the results

4. plt.figure(figsize=(10, 5))
5. plt.subplot(1, 2, 1)
6. plt.imshow(noisy_image, cmap='gray')
7. plt.title('Noisy Image')
8. plt.axis('off')

9. plt.subplot(1, 2, 2)
10. plt.imshow(bilateral_filtered, cmap='gray')
11. plt.title('Bilateral Filtered Image')
12. plt.axis('off')

13. plt.show()
Salt-and-pepper noise
• Salt-and-Pepper Noise, also known as impulse noise, appears
as randomly occurring white (salt) and black (pepper) pixels in
an image.
• This type of noise corrupts pixel intensities and replace them
with maximum or minimum intensity values.

2. Causes of Salt-and-Pepper Noise

1.Transmission Errors: Corruption during image transmission.
2.Faulty Pixels in Sensors: Defective imaging sensors in digital
cameras.
3.Signal Processing Errors: Errors in data conversion or
compression.
1.Effects of Salt-and-Pepper Noise
1.Image Degradation: Loss of fine details and sharpness.
2.Diminished Segmentation Accuracy: Algorithms may fail to detect
edges and features.
3.Reduced Visual Quality: Appearance of white and black spots impacts
visual aesthetics.

Techniques to Reduce Salt-and-Pepper Noise

4.Median Filter: Replaces each pixel value with the median of its
neighbors.
5.Adaptive Filters: Adjust filtering strength based on the local noise level.
6.Non-Local Means Filtering: Considers the similarity between patches
in the image.
Speckle Noise
• Speckle noise is a multiplicative noise that typically
occurs in images captured by coherent imaging
systems, such as radar, ultrasound, and laser-based
imaging devices. Unlike additive noise (e.g., Gaussian
noise), speckle noise is proportional to the intensity of
the image.
Mathematical Model of Speckle
Noise
• Mathemtically, Speckle noise can be represented as:

• Where
• Intensity of the noisy image at pixel location (x, y)
• Intensity of the original image at (x, y)
• Multiplicative noise, which follows a uniform or
Gaussian distributiove around
Causes of Speckle Noise
1. Interference: Constructive and destructive interference of
coherent waves in imaging systems.
2. Scattering: Multiple scattering events in imaging mediums like
tissue in medical imaging.

Characteristics
3. Multiplicative: Noise intensity is proportional to the underlying
image intensity.
4. Grainy Texture: Appears as granular patterns over the image.
5. Spatial Correlation: Noise pattern is spatially correlated,
depending on imaging systems.
Effects
1. Degrades Image Quality: Reduces contrast and obscures details.
2. Hinders Analysis: Affects segmentation, edge detection, and feature
extraction.

Reducing Speckle Noise

• Spatial Domain Techniques:
• Median Filtering: Reduces speckle while preserving edges.
• Adaptive Filtering: Adjusts based on local noise statistics.
• Frequency Domain Techniques:
• Wiener Filtering: Accounts for noise power spectrum.
• Advanced Methods:
• Non-Local Means Filtering, Wavelet Transform, Deep Learning Models
Implementation
Noise Reduction Techniques
(a) Spatial Domain Filtering
1.Mean Filter:
1. Replaces each pixel with the average of its neighborhood.
2. Effective for Gaussian noise but blurs edges.
2.Median Filter:
1. Replaces each pixel with the median of its neighborhood.
2. Excellent for salt-and-pepper noise.
3.Gaussian Filter:
1. Weights neighboring pixels using a Gaussian distribution.
2. Provides smooth results while preserving edges.
(b) Frequency Domain Filtering
1.Low-Pass Filtering:
• Removes high-frequency noise while preserving low-frequency image
features.
2.Band-Stop Filtering:
• Removes specific frequency bands, e.g., periodic noise.
(c) Advanced Noise Reduction Techniques
1. Non-Local Means Denoising:
• Averages similar pixel neighborhoods, regardless of spatial distance.
2. Wavelet Transform Denoising:
• Removes noise in transformed domains like the wavelet domain.
3. Deep Learning Models:
• Convolutional neural networks (CNNs) trained for noise reduction.