MATLAB Image Processing Toolbox

Expert skill for medical image analysis using MATLAB's Image Processing Toolbox (IPT R2025a+).

When to Use This Skill

• Image filtering, denoising, or enhancement in MATLAB
• Image segmentation (thresholding, watershed, active contours, deep learning)
• Morphological operations (erosion, dilation, opening, closing, reconstruction)
• Fourier/frequency domain filtering and transforms
• Image registration and geometric transformations
• Feature extraction (edges, regions, texture)
• Medical imaging: MRI, CT/X-ray, microscopy/histology
• Deep learning segmentation (semanticseg, U-Net, DeepLabv3+)
• Large image processing (blockproc) and GPU acceleration
• Data type handling and conversions

Read Before Coding

Task	Knowledge Card	Key Functions
Reduce noise	filtering-denoising.md	imgaussfilt, medfilt2, wiener2
Detect edges	—	edge, imgradient, imgradientxy
Clean binary masks	morphology-binary.md	bwareaopen, imfill, imopen, imclose
Select threshold	segmentation-thresholding.md	graythresh, imbinarize, multithresh
Separate touching objects	—	watershed, bwdist, imhmin
Measure regions	feature-regions.md	regionprops, bwconncomp, bwlabel
Register images	—	imregister, imregtform, imwarp
Process large images	—	blockproc, gpuArray
Segment with DL	deep-learning-segmentation.md	semanticseg, unetLayers
Convert data types	data-types.md	im2double, im2uint8, mat2gray
Process MRI	medical-mri.md	dicomread, adapthisteq, bias correction
Process CT/X-ray	—	HU windowing, bone segmentation
Count cells	medical-microscopy.md	Cell detection, stain normalization

Note: Rows with "—" indicate functions covered in SKILL.md quick patterns. Future knowledge cards may be added.

Critical Rules

Rule 1: Data Type Handling

The #1 source of errors. Different types have different ranges:

Type	Range	When to Use
uint8	[0, 255]	Display, storage, most IPT functions
uint16	[0, 65535]	Medical images (DICOM), high dynamic range
double	[0, 1] normalized	Arithmetic operations, filtering
logical	0 or 1	Binary masks, segmentation results

% WRONG: Mixed types cause silent truncation
result = uint8_image + double_mask;  % Truncates!

% CORRECT: Explicit conversion
img = im2double(uint8_image);  % [0,255] → [0,1]
result = img + mask;
output = im2uint8(result);     % [0,1] → [0,255]

Rule 2: Filter Normalization

% LOWPASS: Coefficients must sum to 1 (preserves brightness)
h_avg = fspecial('average', 5);       % sum = 1
h_gauss = fspecial('gaussian', 5, 1); % sum ≈ 1

% HIGHPASS: Coefficients must sum to 0 (extracts changes)
h_laplacian = fspecial('laplacian');  % sum = 0
h_log = fspecial('log', 9, 1.5);      % sum = 0

% Verify before custom filters:
assert(abs(sum(h(:)) - 1) < 1e-6, 'Lowpass filter must sum to 1');

Rule 3: Boundary Handling

% For medical images: use 'replicate' to avoid edge artifacts
filtered = imfilter(img, h, 'replicate');

% Options: 'replicate' | 'symmetric' | 'circular' | numeric value
% Default is zero-padding which creates dark borders!

Rule 4: Structuring Element Sizing

SE radius ≈ half the feature size you want to affect.

% Remove noise blobs smaller than ~10 pixels diameter
se = strel('disk', 5);  % radius = 10/2 = 5
cleaned = imopen(bw, se);

% Fill holes up to ~20 pixels diameter
se = strel('disk', 10);
filled = imclose(bw, se);

Rule 5: Threshold Output Range

graythresh returns normalized [0,1] regardless of input type:

img_uint8 = imread('image.png');
level = graythresh(img_uint8);  % Returns 0.45, NOT 115

% For uint8 threshold value:
threshold_uint8 = level * 255;  % 0.45 * 255 = 115

% Better: use imbinarize (handles automatically)
bw = imbinarize(img_uint8);  % Internally uses graythresh

Rule 6: Memory for Large Images

% Use blockproc for images that don't fit in memory
% BorderSize MUST match filter radius to avoid tile artifacts

fun = @(block) imgaussfilt(block.data, 2);
result = blockproc(huge_image, [512 512], fun, ...
    'BorderSize', [6 6], ...      % 3*sigma for Gaussian
    'TrimBorder', true, ...
    'UseParallel', true);

Quick Patterns

Standard Preprocessing Pipeline

function out = preprocess(img)
    img = im2double(img);                           % Normalize
    img = imgaussfilt(img, 1);                      % Denoise
    img = adapthisteq(img, 'ClipLimit', 0.02);      % Enhance contrast
    out = mat2gray(img);                            % Ensure [0,1]
end

Binary Mask Cleanup

function bw = clean_mask(bw, min_area, se_radius)
    se = strel('disk', se_radius);
    bw = imopen(bw, se);              % Remove small protrusions
    bw = imclose(bw, se);             % Close small gaps
    bw = imfill(bw, 'holes');         % Fill holes
    bw = bwareaopen(bw, min_area);    % Remove small objects
end

Segment and Measure

function stats = segment_measure(img)
    bw = imbinarize(img, 'adaptive');
    bw = bwareaopen(bw, 50);
    cc = bwconncomp(bw);
    stats = regionprops('table', cc, img, ...
        'Area', 'Centroid', 'MeanIntensity', 'Eccentricity');
end

Function Quick Reference

Filtering

Function	Purpose	Example
imfilter	Apply custom filter	imfilter(I, h, 'replicate')
imgaussfilt	Gaussian smoothing	imgaussfilt(I, sigma)
medfilt2	Median filter (salt-pepper)	medfilt2(I, [3 3])
wiener2	Adaptive Wiener	wiener2(I, [5 5])
imsharpen	Unsharp masking	imsharpen(I, 'Amount', 1)

Segmentation

Function	Purpose	Example
graythresh	Otsu threshold	level = graythresh(I)
imbinarize	Binarize image	bw = imbinarize(I, 'adaptive')
multithresh	Multi-level threshold	thresh = multithresh(I, 2)
watershed	Watershed segmentation	L = watershed(D)
activecontour	Active contours	bw = activecontour(I, mask)

Morphology

Function	Purpose	Example
strel	Create structuring element	se = strel('disk', 5)
imerode / imdilate	Erosion / Dilation	imerode(bw, se)
imopen / imclose	Opening / Closing	imopen(bw, se)
bwareaopen	Remove small objects	bwareaopen(bw, 100)
imfill	Fill holes	imfill(bw, 'holes')

Features

Function	Purpose	Example
edge	Edge detection	edge(I, 'Canny')
regionprops	Region measurements	regionprops('table', bw, I, 'Area')
bwconncomp	Connected components	cc = bwconncomp(bw)
graycomatrix	Texture GLCM	glcm = graycomatrix(I)

I/O & Types

Function	Purpose	Example
imread / imwrite	Read/write images	I = imread('img.png')
dicomread	Read DICOM	I = dicomread('scan.dcm')
im2double	To double [0,1]	I = im2double(I)
im2uint8	To uint8 [0,255]	I = im2uint8(I)

Knowledge Cards

Detailed documentation organized by topic:

Core Processing:

• knowledge/cards/filtering-denoising.md - Noise reduction techniques
• knowledge/cards/segmentation-thresholding.md - Otsu, adaptive, multi-level
• knowledge/cards/morphology-binary.md - Binary morphological operations

Feature Extraction:

• knowledge/cards/feature-regions.md - regionprops, connected components

Medical Imaging:

• knowledge/cards/medical-mri.md - MRI preprocessing and segmentation
• knowledge/cards/medical-microscopy.md - Cell counting, histology

Advanced:

• knowledge/cards/deep-learning-segmentation.md - semanticseg, U-Net
• knowledge/cards/data-types.md - Type conversions and pitfalls

Cross-Toolbox Integration

For wavelet-based image processing (multiresolution denoising, fusion), see: matlab-wavelet-toolbox skill.

% Example: Wavelet + IPT fusion for MRI denoising
denoised = wdenoise2(mri, 'DenoisingMethod', 'Bayes');  % Wavelet
cleaned = imopen(denoised, strel('disk', 2));           % IPT morphology
enhanced = adapthisteq(cleaned);                        % IPT contrast

---

Source: MathWorks Image Processing Toolbox Documentation (R2025a)