Description
1 Overview
The goal of this lab is to study noise reduction and provide some hands-on experience with fundamental
image enhancement and restoration concepts in the spatial domain.
Image enhancement and restoration are important aspects of image processing, as many real-world applications depend on the underlying quality of an image. Examples include photo enhancement, object
recognition and tracking in surveillance videos, and segmentation of important characteristics (e.g., tumors)
in medical images. For this lab, we will study some fundamental image enhancement and restoration techniques such as noise reduction and image sharpening.
The following images will be used for testing purposes:
• mandrill.png
• cameraman.tif
These images can be downloaded from the course website, and can be loaded using the imread function.
NOTE: The proper way to calculate PSNR is the following (for images normalized to the range 0 to 1):
psnr = 10*log10(1/mean2((f-g).ˆ 2))
2 Noise Generation
To test the effectiveness of an image processing algorithm, it is often necessary to evaluate its performance
under various noise levels to allow for a systematic comparison with other techniques. It is generally not
1
possible to capture real-world images at many noise levels. Furthermore, it is not possible to quantitatively
evaluate the performance of a method when applied to real-world noisy image scenarios since there is no
reference image to compare against. As such, it is often necessary to generate synthetic noise at different
noise levels based on various noise models to evaluate image processing algorithms.
For this study, we will apply additive zero-mean Gaussian (with variance of 0.01), salt and pepper (with
noise density of 0.05), and multiplicative speckle noise (with variance of 0.04) to a synthetic toy image
separately using the imnoise function. The toy image consists of two grayscale bars and can be generated
use the following code
f = [0.3*ones(200,100) 0.7*ones(200,100)];
Plot the noise-contaminated images and the corresponding histograms for each of the noise models.
1. Describe each of the histograms in the context of the corresponding noise models. Why do they appear
that way?
2. Are there visual differences between the noise contaminated images? What are they? Why?
3. In the speckle noise case, what is the underlying distribution used? Can you tell from the histogram?
How?
4. In the speckle noise case, you will notice that the peaks of the histogram are no longer of the same
height as they were in the original image. Also, the spread around each of the peaks is also different
from each other. Why? Hint: Noise is multiplicative.
3 Noise Reduction in the Spatial Domain
Let us now study different noise reduction techniques based on spatial filtering, as well as the effect of
filter parameters on image quality. Load the mandrill image and convert it to a grayscale image using the
rgb2gray function. Furthermore, to get intensity of the image within the range of 0 to 1, use the double
function on the image and then divide it by 255 (alternatively use im2double). To evaluate the noise
reduction performance of various noise reduction techniques, we will contaminate the mandrill image with
zero-mean Gaussian noise with a variance of 0.002. Plot the noisy image and the corresponding histogram
and PSNR between the noisy image and the original noise-free image.
First, let us study the averaging filter method as well as the effect of window size on the noise reduction
performance of a spatial filter. Create a 3×3 averaging filter kernel using the fspecial function. Plot this
filter using imagesc and colormap(gray). Now apply the averaging filter to the noisy image using the
imfilter function. Plot the denoised image and the corresponding histogram. Also, compute the PSNR
between the denoised image and the original noise-free image.
1. Compare the visual difference between the noisy image and the denoised image. How well did it
work? Why? Did the PSNR decrease?
2. Compare the histograms of the noise-free, noisy, and denoised images. What happened? Why?
2
3. Based on visual quality of the denoised image, what are the benefits and drawbacks associated with
the average filter?
Let us now create a 7×7 averaging filter kernel and apply it to the noisy image. Plot the denoised image
and the corresponding histogram. Also, compute the PSNR between the denoised image and the original
noise-free image.
1. Compare the visual difference between the denoised image from the 7×7 filtering kernel and the
denoised image from the 3×3 filtering kernel. Are there any differences? Why? Did the PSNR
decrease? Why?
2. Compare the histograms of the two denoised images. What are the differences? Why?
3. Based on visual quality of the denoised image, what are the benefits and drawbacks associated with
using a larger window size?
Let us now create a 7×7 Gaussian filter kernel with a standard deviation of 1. Plot the filter. Plot the
denoised image and the corresponding histogram. Also, compute the PSNR between the denoised image
and the original noise-free image.
1. Compare the visual difference between the denoised image from the Gaussian filtering kernel and the
denoised images from the averaging filter kernels. Are there any differences? Why? Did the PSNR
decrease? Why?
2. Compare the histograms of the denoised image using the Gaussian filtering kernel and the denoised
images from the averaging filter kernels. What are the differences? Why?
3. Based on visual quality of the denoised image, what are the benefits and drawbacks associated with
using a Gaussian kernel as opposed to an averaging kernel?
Let us now create a new noisy image by adding salt and pepper noise to the image. Apply the 7×7 averaging
filter and the Gaussian filter to the noisy image separately. Plot the noisy image, the denoised images using
each method, and the corresponding histograms. Also, compute the PSNR between the denoised images
and the original noise-free image.
1. How does the averaging filter and Gaussian filtering methods perform on the noisy image in terms of
noise reduction? Explain in terms of visual quality as well as PSNR. Why do we get such results?
2. Compare the histograms of the denoised images with that of the noisy image. What characteristics
are present in all of the histograms? Why?
Let us now apply the median filter on the noisy image. The medfilt2 function will come in handy for this.
Plot the denoised image and the corresponding histogram. Also, compute the PSNR between the denoised
image and the original noise-free image.
3
1. How does the denoised image produced using the median filter compare with the denoised images
produced using averaging filter and Gaussian filtering methods? Explain in terms of visual quality
as well as PSNR. Why do we get such results with median filter when compared to the other spatial
filtering methods?
4 Sharpening in the Spatial Domain
Let us now briefly study sharpening techniques based on spatial filtering as well as the effect of sharpening
filter parameters on image quality. Load the Cameraman image and get intensity of the image within the
range of 0 to 1. One very useful and customizable technique for sharpening images is high-boost filter. Let
us study it at its various stages. First, apply the 7×7 Gaussian filter on the Cameraman image and subtract
the Gaussian-filtered image from the original Cameraman image. Plot both the Gaussian-filtered image and
the subtracted image.
1. What does the subtracted image look like? What frequency components from the original image are
preserved in the subtracted image? Why?
Now we add the subtracted image to the original image. Plot the resulting image.
1. What does the resulting image look like? How does it differ from the original image? Explain why it
appears this way.
Now, instead of adding the subtracted image to the original image, multiply the subtracted image by 0.5 and
then add it to the original image. Plot the resulting image.
1. Compare the results produced by adding the subtracted image to the original image and that produced
by adding half of the subtracted image to the original image. How does it differ? Explain why it
appears this way.
2. What does multiplying the subtracted image by a factor less than one accomplish? What about greater
than one?
5 Report
Include in your report:
• A brief introduction.
• Pertinent graphs and images (properly labelled).
• Include code (can be included in appendix).
• Include responses to all questions.
• A brief summary of your results with conclusions.
4
