Matlab中图像边缘检测功能失效问题排查求助

Hey there, let's dig into why your edge detection script isn't working as expected. I spotted a few key issues in your code that are keeping it from producing the edge results you want. Let's break them down one by one:

1. Incorrect intensity range for double-precision images #

When you convert the grayscale image to double directly with c = double(c);, pixel values stay in the 0-255 range. But functions like imnoise and imshow are designed to work with double images in the 0-1 range by default. This mismatch leads to noisy gradients and incorrect visualizations.

2. Convolution boundary padding causes size and display issues #

The default conv2 uses 'full' padding, which makes the output gradient matrices (grad_g, grad_h, grad) larger than the original image. While your loop uses the size of grad, the bigger problem is raw gradient values will be way outside the 0-1 range, so imshow can't render them properly.

3. Incomplete imshow calls and improper thresholded image display #

Your imshow(... line is cut off, and even if completed, displaying the raw grad matrix (with large values) or the 0/1 thresh_grad matrix directly won't show edges as intended—Matlab expects double images to be in 0-1, or requires explicit display range scaling.

Here's the fixed version of your script with detailed explanations for each change:

clear all; close all; clc;

% Load and preprocess the image correctly
c = rgb2gray(imread('image_S004_I0004.jpg'));
c = im2double(c); % Convert to standard 0-1 double range instead of 0-255

% Add salt & pepper noise (works correctly with 0-1 double input)
k = imnoise(c, 'salt & pepper', 0.01);

% Sobel edge detection kernels
gg = [-1 0 1;-2 0 2; -1 0 1]; % Horizontal Sobel kernel
gh = gg'; % Vertical Sobel kernel

% Use 'same' padding to keep gradient size matching original image
grad_g = conv2(k, gg, 'same');
grad_h = conv2(k, gh, 'same');

% Compute gradient magnitude and normalize for display
grad = sqrt(grad_g.^2 + grad_h.^2);
grad_normalized = mat2gray(grad); % Scale gradient values to 0-1 range

% Thresholding (adjusted to match 0-1 intensity range)
T = 80/255; % Equivalent to original 80 threshold for 0-255 range
[r, s] = size(grad_normalized);
thresh_grad = zeros(r, s); % Pre-allocate matrix for efficiency
for ii = 1:r
    for jj = 1:s
        if grad_normalized(ii, jj) < T
            thresh_grad(ii, jj) = 0;
        else
            thresh_grad(ii, jj) = 1;
        end
    end
end

% Display results properly
figure()
subplot(121); imshow(grad_normalized); title('Gradient Magnitude');
subplot(122); imshow(thresh_grad); title('Thresholded Edges');

Key fixes explained:

• im2double instead of double: Converts the uint8 image directly to a 0-1 double matrix, which is Matlab's standard for image processing operations.
• 'same' padding in conv2: Ensures gradient matrices match the input image size, eliminating unexpected extra pixels around edges.
• mat2gray for gradient normalization: Scales gradient values to 0-1 so imshow renders them correctly without clipping.
• Adjusted threshold value: 80/255 aligns the threshold with the 0-1 intensity range, matching your original intent.
• Pre-allocated thresh_grad: A small Matlab best practice to speed up loops, especially for larger images.
• Completed imshow calls: Added titles and proper rendering for both the gradient magnitude and thresholded edges.

内容的提问来源于stack exchange，提问作者Bacon