MATLAB - Laplacian Filter

Modern Image Filtering in MATLAB: The Laplacian Filter

In digital image processing, filtering is a fundamental technique for enhancement and feature extraction. The Laplacian filter is a second-order derivative filter used to highlight regions of rapid intensity change. This makes it highly effective for edge detection and image sharpening.

This updated tutorial explores how to apply the Laplacian filter in modern MATLAB, focusing on best practices, practical applications like image sharpening, and understanding the underlying concepts.

Core Concept: The Laplacian Operator

The Laplacian is a 2D measure of the second spatial derivative of an image. Unlike first-order derivatives (like Sobel or Prewitt filters) which detect the presence of an edge, the second-order derivative is zero on a ramp edge and produces a ‘double edge’ at a step edge. Its key feature is its ability to detect whether a pixel is on the lighter or darker side of an edge, making it excellent for finding fine details and for sharpening.

It’s typically implemented as a convolution with a kernel (or mask). Two common kernels are:

% 4-connectivity kernel
kernel_4_conn = [ 0 -1  0;
               -1  4 -1;
                0 -1  0];

% 8-connectivity kernel (includes diagonals)
kernel_8_conn = [-1 -1 -1;
               -1  8 -1;
               -1 -1 -1];

Method 1: The High-Level imfilter Function (Recommended)

The imfilter function is the recommended high-level tool for linear and non-linear spatial filtering in MATLAB. It handles data types, boundary conditions, and color channels automatically, making it robust and easy to use.

Syntax

filteredImage = imfilter(inputImage, kernel, boundaryOptions, ...);

• inputImage: The image to be filtered (grayscale or RGB).
• kernel: The filter kernel, such as the Laplacian kernel defined above.
• boundaryOptions (Optional): A string specifying how to handle image boundaries, e.g., 'replicate' (default), 'symmetric', or `‘circular’.

Example 1: Edge Detection with imfilter

This example demonstrates applying the Laplacian filter to detect edges. Note that the raw output highlights edges but isn’t typically used as a final visual result.

% 1. Read a sample image
originalImage = imread('cameraman.tif');

% Best practice: Convert image to double for precise calculations
% The range [0, 1] is standard for floating-point images.
originalImage_double = im2double(originalImage);

% 2. Define the Laplacian kernel
laplacianKernel = [0 -1 0; -1 4 -1; 0 -1 0];

% 3. Apply the filter
laplacianOutput = imfilter(originalImage_double, laplacianKernel, 'replicate');

% 4. Display the results
figure;
subplot(1, 2, 1);
imshow(originalImage_double);
title('Original Image');

% The output has positive and negative values. The `[]` argument to imshow
% scales the display range to show the full detail.
subplot(1, 2, 2);
imshow(laplacianOutput, []);
title('Laplacian Filter Output');

Practical Application: Image Sharpening

A primary use of the Laplacian filter is to sharpen an image. This is achieved by subtracting the Laplacian-filtered image (which represents the edges) from the original image. This enhances the edges, making the image appear sharper.

% Continuing from the previous example...

% Sharpen the image by subtracting the Laplacian result
% The scaling factor (e.g., 1.0) can be adjusted to control sharpening strength.
sharpeningFactor = 1.0;
sharpenedImage = originalImage_double - sharpeningFactor * laplacianOutput;

% Display the sharpened image
figure;
subplot(1, 2, 1);
imshow(originalImage_double);
title('Original Image');

subplot(1, 2, 2);
imshow(sharpenedImage);
title('Sharpened Image');

Method 2: The Fundamental conv2 Function

conv2 performs 2D convolution and is the underlying mathematical operation of imfilter. Using conv2 is more fundamental and can be useful for educational purposes or when custom behavior is needed, but it requires more manual handling.

Syntax

C = conv2(A, B, 'shape');

• A: The input matrix (image). Must be a floating-point type like double.
• B: The 2D filter kernel.
• 'shape': An optional parameter. 'same' is typically used to ensure the output has the same size as the input matrix A.

Example 2: Filtering with conv2

% 1. Read and convert the image
img_gray = imread('peppers.png');

% `conv2` requires a double-precision input. If the input is RGB,
% convert it to grayscale first.
if size(img_gray, 3) == 3
    img_gray_double = im2double(rgb2gray(img_gray));
else
    img_gray_double = im2double(img_gray);
end

% 2. Define the Laplacian filter kernel
laplacianKernel = [-1 -1 -1; -1 8 -1; -1 -1 -1]; % 8-connectivity

% 3. Apply the Laplacian filter using conv2
filtered_img_conv = conv2(img_gray_double, laplacianKernel, 'same');

% 4. Display the original and filtered images
figure;
subplot(1, 2, 1);
imshow(img_gray_double);
title('Original Grayscale Image');

subplot(1, 2, 2);
imshow(filtered_img_conv, []);
title('Laplacian Filtered (conv2)');

Modern Best Practices & Common Errors

• Data Types: Always convert images to a floating-point type (double or single) before performing filtering to avoid rounding errors and integer overflow. Use im2double() or im2single().
• Color Images: imfilter handles RGB images automatically by filtering each channel separately. If using conv2, you must manually loop through the color channels.
• Laplacian of Gaussian (LoG): The Laplacian is very sensitive to noise. A common practice is to first smooth the image with a Gaussian filter and then apply the Laplacian. This combined operation is called the Laplacian of Gaussian (LoG) filter. You can create one easily with fspecial('log', [size], sigma).
• Visualization: The output of a Laplacian filter contains both positive and negative values and is not a viewable image in the standard [0, 255] or [0, 1] range. Use imshow(output, []) to scale the display range and visualize the result correctly.