Morphological Transformation¶

v2¶

Apply morphological transformations to color images using configurable operations to modify object shapes, remove noise, fill holes, detect edges, and extract features. Color images are processed with transformations applied to all channels, and the output is returned as color for compatibility with downstream processing blocks.

How This Block Works¶

This block applies morphological transformations directly to color images (BGR or BGRA). The block:

1. Receives an input image to transform (works with color or grayscale images)
2. Ensures image is in color format (if grayscale, converts to BGR for consistent color output)
3. Creates a square structuring element (kernel) of specified kernel_size
4. Applies the selected morphological operation to all channels simultaneously
5. Preserves the color format throughout the transformation (including alpha channel for BGRA images)
6. Returns the morphologically transformed color image

Supported operations:

• Erosion: Shrinks objects by removing boundary pixels (separates objects, removes noise)
• Dilation: Expands objects by adding boundary pixels (connects objects, fills holes)
• Opening: Erosion followed by dilation (removes noise, separates objects)
• Closing: Dilation followed by erosion (fills holes, connects fragments)
• Opening then Closing: Opens then closes (specialized preprocessing for edge detection and refinement)
• Gradient: Dilation minus erosion (highlights object boundaries/edges)
• Top Hat: Input minus opening (detects small bright details)
• Black Hat: Closing minus input (detects small dark details)

All operations preserve the color format, making output compatible with downstream color-based blocks like Mask Edge Snap.

Common Use Cases¶

• Color Image Preprocessing: Process color images while preserving color information for downstream blocks
• Edge Detection and Refinement: Use opening+closing for specialized preprocessing before edge snapping
• Noise Removal: Remove noise from color images while maintaining color channels
• Feature Extraction: Extract morphological features from color images
• Image Conditioning: Prepare color images for color-based downstream processing

Connecting to Other Blocks¶

• Before Mask Edge Snap: Output color images for edge snapping and mask refinement
• Before visualization blocks: Display morphologically processed color images
• In color image pipelines: Process color images while maintaining color information

Type identifier¶

Use the following identifier in step "type" field: roboflow_core/morphological_transformation@v2to add the block as as step in your workflow.

Properties¶

The Refs column marks possibility to parametrise the property with dynamic values available in workflow runtime. See Bindings for more info.

Available Connections¶

Input and Output Bindings¶

The available connections depend on its binding kinds. Check what binding kinds Morphological Transformation in version v2 has.

{
    "name": "<your_step_name_here>",
    "type": "roboflow_core/morphological_transformation@v2",
    "image": "$inputs.image",
    "kernel_size": "5",
    "operation": "Closing"
}

v1¶

Apply morphological transformations to images using configurable operations (erosion, dilation, opening, closing, gradient, top hat, black hat) to modify object shapes, remove noise, fill holes, detect edges, and extract features from binary or grayscale images for image processing, noise removal, and feature extraction workflows.

How This Block Works¶

This block applies morphological transformations to images using structuring elements (kernels) to modify object shapes and extract features. The block:

1. Receives an input image to transform (color images are automatically converted to grayscale)
2. Converts the image to grayscale if it's in color format (morphological operations work on single-channel images)
3. Creates a square structuring element (kernel) of specified kernel_size:
4. Creates a square matrix of ones with dimensions (kernel_size, kernel_size)
5. The kernel defines the neighborhood used for morphological operations
6. Larger kernels affect larger regions, smaller kernels affect smaller regions
7. Applies the selected morphological operation:

For Erosion: - Shrinks objects by removing pixels from object boundaries - Replaces each pixel with the minimum value in its kernel neighborhood - Thins objects and removes small protrusions - Useful for separating touching objects and removing small noise

For Dilation: - Expands objects by adding pixels to object boundaries - Replaces each pixel with the maximum value in its kernel neighborhood - Thickens objects and fills small holes - Useful for connecting nearby objects and filling gaps

For Opening (Erosion followed by Dilation): - First erodes then dilates the image - Removes small noise and separates touching objects - Preserves larger objects while eliminating small details - Useful for noise removal and object separation

For Closing (Dilation followed by Erosion): - First dilates then erodes the image - Fills small holes and connects nearby objects - Preserves object shape while closing gaps - Useful for filling holes and connecting fragmented objects

For Gradient (Dilation minus Erosion): - Computes the difference between dilated and eroded images - Highlights object boundaries and edges - Creates an outline effect showing object perimeters - Useful for edge detection and boundary extraction

For Top Hat (Input minus Opening): - Computes the difference between original image and its opening - Highlights small bright details that were removed by opening - Emphasizes bright features smaller than the kernel - Useful for detecting small bright objects or details

For Black Hat (Closing minus Input): - Computes the difference between closing and original image - Highlights small dark details that were filled by closing - Emphasizes dark features smaller than the kernel - Useful for detecting small dark objects or holes

1. Preserves the image structure and metadata
2. Returns the morphologically transformed image

Morphological operations use structuring elements (kernels) to probe and modify image structures. The kernel_size controls the scale of the transformation - larger kernels affect larger regions and have more pronounced effects. Basic operations (erosion, dilation) modify object sizes, composite operations (opening, closing) combine effects for noise removal and gap filling, and derived operations (gradient, top hat, black hat) extract specific features. The operations work best on binary or high-contrast grayscale images where object boundaries are clearly defined.

Common Use Cases¶

• Noise Removal: Remove small noise and artifacts from binary or grayscale images (e.g., remove salt-and-pepper noise, eliminate small artifacts, clean up thresholded images), enabling noise removal workflows
• Object Separation: Separate touching or overlapping objects in images (e.g., separate touching objects, split connected regions, isolate individual objects), enabling object separation workflows
• Hole Filling: Fill small holes and gaps in objects (e.g., fill holes in objects, close gaps in shapes, complete fragmented objects), enabling hole filling workflows
• Edge Detection: Detect object boundaries and edges using morphological gradient (e.g., find object edges, extract boundaries, detect object outlines), enabling morphological edge detection workflows
• Feature Extraction: Extract specific features like small bright or dark objects (e.g., detect small bright details with top hat, find small dark objects with black hat, extract specific morphological features), enabling feature extraction workflows
• Image Preprocessing: Prepare binary or grayscale images for further processing (e.g., clean up thresholded images, prepare images for contour detection, preprocess images for analysis), enabling morphological preprocessing workflows

Connecting to Other Blocks¶

This block receives an image and produces a morphologically transformed image:

• After image thresholding blocks to clean up thresholded binary images (e.g., remove noise from thresholded images, fill holes in binary images, separate objects in thresholded images), enabling thresholding-to-morphology workflows
• After preprocessing blocks to apply morphological operations after other preprocessing (e.g., apply morphology after filtering, clean up after thresholding, process after enhancement), enabling multi-stage preprocessing workflows
• Before contour detection blocks to prepare images for contour detection (e.g., clean up images before contour detection, fill holes before finding contours, separate objects before contour analysis), enabling morphology-to-contour workflows
• Before analysis blocks that process binary or grayscale images (e.g., analyze cleaned images, process separated objects, work with morphologically processed images), enabling morphological analysis workflows
• Before visualization blocks to display morphologically transformed images (e.g., visualize cleaned images, display processed results, show transformation effects), enabling morphological visualization workflows
• In image processing pipelines where morphological operations are part of a larger processing chain (e.g., clean images in pipelines, apply morphology in multi-stage workflows, process images in transformation chains), enabling morphological processing pipeline workflows

Requirements¶

This block works on single-channel (grayscale) images. Color images are automatically converted to grayscale before processing. Morphological operations work best on binary or high-contrast grayscale images where object boundaries are clearly defined. For optimal results, use thresholded binary images or high-contrast grayscale images as input.

Type identifier¶

Use the following identifier in step "type" field: roboflow_core/morphological_transformation@v1to add the block as as step in your workflow.

Properties¶

The Refs column marks possibility to parametrise the property with dynamic values available in workflow runtime. See Bindings for more info.

Available Connections¶

Input and Output Bindings¶

The available connections depend on its binding kinds. Check what binding kinds Morphological Transformation in version v1 has.

{
    "name": "<your_step_name_here>",
    "type": "roboflow_core/morphological_transformation@v1",
    "image": "$inputs.image",
    "kernel_size": "5",
    "operation": "Closing"
}