University of the West of England

Designing a spatial filter to sharpen an image

UFMFH8-15-3 Digital Signal Processing

Student Name : HPRS Siriwardana

Module Leader : Ms. Maithrayei Sivalingam
 CONTENT

01. Introduction

02. Image input

03. Kernel design

04. Filter Implementation

05. Sharpening Operation

06. Performance Evaluation

07. Optimization and fine-tuning

 01 INTRODUCTION

Image sharpening plays a crucial role in various fields, including

photography, medical imaging, and computer vision, where enhancing edges and
fine details can significantly improve visual quality and aid in analysis. Spatial
filters, such as the Laplacian filter, offer a simple yet effective method for
sharpening images by highlighting areas of rapid intensity change. This process
involves convolving the image with a specific filter kernel, which accentuates
edges and enhances contrast. In this discussion, we explore the design and
application of a spatial filter, focusing on the Laplacian filter, to sharpen digital
images and improve their overall clarity and detail.
 02.IMAGE INPUT

Image Loading and Representation

MATLAB's image processing toolbox offers various functions for loading images in different
formats such as JPEG, PNG, BMP, etc. The imread function is the primary tool for this task. It
reads the image file specified by its path and stores it as a matrix of pixel values. Grayscale
images are represented as 2D matrices, while color images are stored as 3D arrays, with each
dimension representing the image's height, width, and color channels respectively.
Understanding the image's data type and dimensions is crucial for subsequent processing steps.

Image Preprocessing

Preprocessing steps are often necessary before applying filters or conducting other operations on
images. These steps might include noise reduction, resizing, color space conversion, or contrast
enhancement. MATLAB provides a plethora of functions for these tasks, such as imgaussfilt
for Gaussian blur, imresize for resizing, rgb2gray for converting color images to grayscale,
and more. Proper preprocessing can significantly improve the effectiveness of subsequent image
processing algorithms.

Handling Grayscale and Color Images

Distinguishing between grayscale and color images is essential for selecting appropriate
processing techniques. Grayscale images have a single channel representing pixel intensities,
whereas color images consist of multiple channels (typically red, green, and blue). MATLAB's
size function can be employed to inspect the dimensions of the loaded image matrix, facilitating
the determination of the number of channels.

Image Visualization

Visualizing images is crucial for understanding their contents and the effects of processing
operations. MATLAB's imshow function is commonly used to display images in a figure
window. This allows for the visual inspection of both original and processed images, aiding in
debugging and fine-tuning of image processing algorithms.

Saving Processed Images:

After processing, it is often necessary to save the results for further analysis or sharing.
MATLAB's imwrite function enables the saving of images in various file formats, allowing
users to specify the output file path and format. This functionality facilitates the integration of
MATLAB-based image processing pipelines into larger workflows.

By mastering these aspects of image input handling in MATLAB, users can develop robust and
flexible image processing pipelines capable of handling diverse input scenarios effectively.
 03 KERNEL DESIGN

Kernel Design

The kernel, also known as a filter or mask, plays a crucial role in image processing operations
such as sharpening. It defines the local operation applied to each pixel in the image. Different
kernels yield different effects, so selecting an appropriate kernel is essential for achieving the
desired results.

Common Kernel Types

1. Laplacian Kernel: Often used for edge detection and image sharpening, the Laplacian
kernel emphasizes high-frequency components in the image. It highlights abrupt intensity
changes, such as edges.
2. Sobel and Prewitt Kernels: These kernels are widely used for edge detection. They
approximate the gradient of the image intensity, making them effective for highlighting
edges in different directions.
3. Gaussian Kernel: Used for blurring and noise reduction, the Gaussian kernel applies a
weighted average to each pixel's neighborhood, with weights determined by a Gaussian
distribution. It effectively smooths the image while preserving edges.

Kernel Size and Shape

The size and shape of the kernel determine the extent of the local operation and its effect on the
image. Larger kernels capture more spatial information but may introduce more blurring.
Common kernel sizes range from 3x3 to 7x7, but larger sizes can be used for specific
applications.

Kernel Coefficients

The coefficients of the kernel determine the strength and direction of the operation applied to
each pixel. For example, in the Laplacian kernel, the central coefficient is often positive,
surrounded by negative coefficients to emphasize intensity changes.

Custom Kernel Design

MATLAB allows users to design custom kernels tailored to specific image processing tasks.
These kernels can be created manually or generated algorithmically based on desired
characteristics. Users can experiment with different coefficients and kernel shapes to achieve
optimal results for their applications.
Kernel Visualization

Visualizing kernels can provide insights into their effects on image processing. MATLAB's
plotting functions can be used to visualize kernel matrices as heatmaps or surface plots, aiding in
understanding their spatial characteristics and effects on image features.

04 FILTER IMPLEMENTATION

Filter Implementation

In MATLAB, implementing filters involves convolving the image with a kernel. Convolution is
a fundamental operation in image processing that applies a filter to an image by sliding the
kernel over the image's pixels and computing a weighted sum of neighboring pixel values.

Convolution Function

MATLAB provides the conv2 function for performing 2D convolution, which is suitable for
image processing tasks. The function takes the input image and the kernel as inputs and returns
the filtered image.

Handling Boundary Effects

During convolution, the kernel extends beyond the boundaries of the image. MATLAB offers
different options for handling boundary effects, such as zero-padding, symmetric padding, or
circular padding. The choice of padding method can affect the results, especially near the image
edges.

Filtering Modes

The conv2 function supports different filtering modes, including 'full', 'same', and 'valid'.

• 'Full' mode returns the full convolution, resulting in an output image larger than the input.
• 'Same' mode returns an output image with the same size as the input by padding the edges
as necessary.
• 'Valid' mode computes only the parts of the convolution where the kernel and the image
overlap, resulting in an output image smaller than the input.

Performance Considerations

Efficient implementation of filters is crucial, especially for large images or complex kernels.
MATLAB offers various optimization techniques to enhance performance, such as vectorization,
parallel processing, and utilizing built-in functions optimized for specific operations.
Custom Filter Implementation

MATLAB allows users to define custom filtering operations by manually iterating over image
pixels and computing the weighted sum using nested loops. While this approach offers
flexibility, it may not be as efficient as using built-in functions for convolution.

Visualizing Filtered Images

After applying the filter, visualizing the filtered images helps assess the effectiveness of the
filtering operation. MATLAB's imshow function can display both the original and filtered
images, enabling visual inspection of the image enhancement results.

05 SHARPENING OPERATION

Sharpening Operation

After applying the filter to the input image, the sharpening operation enhances the edges and
fine details to make them more pronounced. This operation aims to improve image clarity and
emphasize important features.

Enhancement Technique

In MATLAB, the sharpening operation is typically achieved by adding the filtered image back
to the original image, with appropriate scaling. This process increases the contrast along edges,
making them appear sharper.

Adjusting Strength

The strength of the sharpening effect can be adjusted by scaling the filtered image before adding
it to the original image. This scaling factor determines the degree of enhancement applied to the
image. Higher scaling factors result in stronger sharpening effects but may also introduce
artifacts or noise amplification.

Alternative Techniques

Alternatively, sharpening can be performed by subtracting the filtered image from the original
image. This technique highlights the high-frequency components in the image, effectively
enhancing edges and details.
Clipping Values

After applying the sharpening operation, it's essential to clip the pixel values to ensure they
remain within the valid range. In MATLAB, pixel values are typically normalized to the range
[0, 1] before displaying or saving the image.

06 PERFORMANCE EVALUATION

Performance Evaluation

Assessing the effectiveness of image processing algorithms, including sharpening filters,

requires quantitative and qualitative evaluation methods. These methods help measure the
fidelity of the processed images compared to the originals and provide insights into the
algorithm's performance.

Quantitative Metrics

MATLAB provides various quantitative metrics for evaluating image quality. Some commonly
used metrics include:

• Peak Signal-to-Noise Ratio (PSNR): Measures the ratio between the maximum possible
signal power and the power of corrupting noise. Higher PSNR values indicate better
image quality.
• Mean Squared Error (MSE): Calculates the average squared difference between
corresponding pixels in the original and processed images. Lower MSE values indicate
better image quality.
• Structural Similarity Index (SSIM): Computes a measure of similarity between the
original and processed images based on luminance, contrast, and structure. SSIM values
range from -1 to 1, with 1 indicating perfect similarity.

Visual Inspection

While quantitative metrics provide objective measures of image quality, visual inspection by
human observers is crucial for assessing perceptual quality. MATLAB's imshow function can be
used to display images side by side for visual comparison. Observers can provide subjective
feedback on the sharpness, clarity, and overall quality of the processed images.
Automated Testing

For large-scale evaluations or comparative studies, MATLAB's scripting capabilities enable

automated testing of image processing algorithms. Test scripts can apply the sharpening filter to
a dataset of images, compute quantitative metrics, and generate reports summarizing the results.

Cross-Validation

Cross-validation techniques can be employed to assess the generalization performance of

sharpening filters. MATLAB provides functions for splitting datasets into training and testing
sets, enabling rigorous evaluation of algorithm performance on unseen data.

Parameter Tuning

Fine-tuning the parameters of the sharpening filter based on performance evaluation results is
essential for optimizing its effectiveness. MATLAB's optimization toolbox offers tools for
parameter optimization, such as grid search, random search, and gradient-based optimization
algorithms.

Validation with Ground Truth

When available, ground truth images can be used for validation purposes. MATLAB facilitates
the comparison of processed images with ground truth images, enabling precise evaluation of the
sharpening filter's performance.

07 Optimization and fine-tuning

Optimization and Fine-Tuning

Optimizing and fine-tuning image processing algorithms, such as sharpening filters, is essential
for improving their effectiveness and efficiency. MATLAB offers various tools and techniques
for optimizing algorithm parameters and enhancing performance.
Parameter Optimization

Adjusting the parameters of the sharpening filter can significantly impact its performance.
MATLAB provides optimization techniques such as grid search, random search, and gradient-
based optimization algorithms to find the optimal parameter values. These techniques iteratively
explore the parameter space to maximize performance metrics such as PSNR, SSIM, or
subjective quality.

Performance Evaluation Feedback

Feedback from performance evaluation metrics and visual inspection guides the optimization
process. MATLAB allows users to integrate performance evaluation scripts into the optimization
workflow, enabling automated parameter tuning based on evaluation results. This iterative
process helps refine the algorithm's parameters for optimal performance across different image
datasets.

MATHLAB CODE

originalImage = imread('C:\Users\Acer\Desktop\sunethbiet\12.png');
originalImage = im2double(originalImage);
gaussianSigma = 2;
sharpeningAmount = 1.5;
gaussianFilter = fspecial('gaussian', [5 5], gaussianSigma);
smoothedImage = imfilter(originalImage, gaussianFilter, 'replicate');
detailImage = originalImage - smoothedImage;
sharpenedImage = originalImage + sharpeningAmount * detailImage;
sharpenedImage = max(0, min(sharpenedImage, 1));
figure;
subplot(1, 2, 1);
imshow(originalImage);
title('Original Image');
subplot(1, 2, 2);
imshow(sharpenedImage);
title('Sharpened Image');
Sharpen image by mathlab