Automatic Segmentation

Difference in Job Mode

The available functionality of the individual steps can differ between the wizard in Free Mode and the steps that can be added in Job Mode, e.g. the Interactive checkbox is not visible in Job Mode.

In this step you can select the segmentation method that is applied and set parameters for the segmentation of the objects that you want to measure. All the objects detected with the current settings are highlighted in the image. Note that during the setup of the analysis (via Setup Image Analysis in Free Mode), the segmentation is only performed on the area visible in the viewport. If you enter the interactive analysis or are in Job Mode, the image will be fully segmented.

Parameter		Description
Execute		Activated: This step is included when the analysis is run. Otherwise the step is skipped.
Interactive		Activated: The segmentation can be changed interactively while the analysis setting is run.
Class List		Selects the class for which you want to define the segmentation. You can specify different settings for each class.
Segmentation Method		Displays and selects a segmentation method for the currently selected class with the dropdown.
-	Explore	Opens the dialog to explore and select an available segmentation method, see Segmentation Method Selection Dialog.

The visible parameters depend on the selected segmentation method. The following parameters sections can be available:

Smoothing Section

Parameter		Description
Smoothing		Selects how to smooth the image before the threshold values are set. The following methods are available:
-	None	The image is not smoothed.
-	Lowpass	Applies the Lowpass method. The lowpass filter compares the brightness of each pixel to the brightness of its neighboring pixels. If a pixel is brighter than its neighbors, the brightness of this pixel is reduced and the brightness of the neighboring pixels is increased. This suppresses sharp changes in brightness (i.e. contours) and leads to more gradual changes in brightness.
-	Gauss	Applies the Gauss method. Each pixel is replaced by a weighted average of its neighbors. The weighting depends on the sigma value. The Gaussian filter is particularly useful for contour enhancement, which is very sensitive to noise. Using a Gaussian filter before finding contours greatly improves the results.
-	Median	Applies the Median method. Each pixel is replaced by the median of its neighbors. The number of neighboring pixels taken into account depends on the size. In a set of values (in this case the pixel values taken into account), the median is the value for which the number of larger values is equal to the number of smaller values.
Size		Only visible, if you have selected Low Pass or Median. Sets the size of the filter matrix in the X and Y direction, i.e. the number of neighboring pixels taken into account. The size should correspond to the pixel size of the contours to be reduced.
Sigma		Only visible, if you have selected Gauss. Sets the sigma value that defines how much neighboring pixels contribute to the weighting. Larger values broaden the applied Gaussian distribution and lead to reduced noise but also to an increased loss of image information.

Sharpen Section

Parameter		Description
Sharpen		Select how to improve the sharpness by enhancing contrast at fine structures and edges of the image before the threshold values are set. The following methods are available:
-	None	No sharpening algorithm is applied.
-	Delineate	Applies the Delineate method. It emphasizes edges around structures in an image, which is useful for images where the gray value range of structures differs clearly from the gray value range of the pixels around them.
-	Unsharp Masking	Applies the Unsharp Masking method.
Threshold		Only visible, if you have selected Delineate. Sets the threshold value for edge detection. The threshold value should correspond roughly to the gray value difference between objects and the background.
Size		Only visible, if you have selected Delineate. Sets the size of the edge detection filter, i.e. the size of image details which are enhanced. The smaller the Size value is, the finer are the details affected by the tool. The value should correspond to the size of the transition area between objects and the background.
Strength		Only visible, if you have selected Unsharp Masking. Sets the strength of the Unsharp Masking. The higher the value selected, the greater the extent to which small structures are enhanced.

Variance Section

This section is only visible if Variance-Based Thresholding is selected.

Parameter		Description
Kernel Size		Sets the kernel size used to calculate the variance value of one pixel with its neighboring pixels.
Variance		Defines the lower and upper threshold for the variance.

Subtract BG Section

Parameter		Description
Subtract BG		Only visible if Segmentation with Background Subtraction is selected. Selects which kind of background subtraction is performed.
-	None	No background subtraction is performed.
-	Rolling ball	The rolling ball background subtraction is performed.

Object Size & Hole Section

Parameter		Description
Min. Object Size		Sets the minimum size in pixels that an object must have in order to be segmented.
Min. Hole Size		Sets the minimum size in pixels that a hole must have in order to be recognized for segmentation. This input is synchronized with the input for Min. Object Size, which must not be smaller than Min. Hole Size.
Fill all Holes		Specifies how holes in detected objects are treated.
–	On	Fills holes in segmented objects ().
–	Off	Does not fill the holes in segmented objects ().

Binary Section

Parameter		Description
Binary		Selects which morphological operations are performed on the segmented (binary) image.
-	None	No operation is performed.
-	Open	Performs first erosion and then dilation. The effect is smoothing and removing of isolated pixels.
-	Close	Performs first dilation and then erosion. The effect is smoothing of the objects and filling of small holes.
-	Dilate	Enlarges the boundaries of segmented regions. Areas grow in size and holes within the regions become smaller.
-	Erode	Erodes boundaries of the segmented regions. The areas shrink in size and holes within the areas become larger.
Count		Sets how often the selected binary operation is performed with the slider or input field.

Separate Section

Parameter		Description
Separate		Selects whether you want to process the image further after segmentation. Objects that are touching one another can be separated using different methods.
-	None	Objects are not separated.
-	Morphology	Separates objects by first reducing and then enlarging them, making sure that once objects have been separated they do not merge together again.
-	Watersheds	Separates objects that are roughly the same shape. The result is two shapes separated by a thin 1-pixel boundary. The rest of the shape perimeter remains unchanged. This method may however result in the splitting of elongated objects.
Count		Sets the count value, which is similar to a Sigma for Gauss applied to a binary image.

Suppress Section

Parameter	Description
Suppress Invalid	Activated: Discards invalid pixels at the border of the image.
Suppress Border	Only visible if Variance-Based Thresholding is selected. Activated: Suppresses the border pixels which might be incorrect, as areas outside of the image are filled with zeros. The excluded area depends on the used kernel size.

Segmentation Method Selection Dialog

With this dialog you can select the method that is used for segmenting the class currently selected in the class list.

Parameter		Description
Global Thresholding		Applies a fixed threshold across the entire image to segment objects of interest. Use it for images with consistent object intensity like fluorescence images.
AI Instance Segmentation		Uses a deep learning model trained on arivis Cloud to distinguish and separate individual objects, even if touching or overlapping. Use it for precise identification in complex images. You need the Docker Desktop software running on your PC and have a suitable model available, see Machine Learning and AI Functionalities in ZEN core.
AI Semantic Segmentation		Uses a semantic machine learning model trained with Intellesis in ZEN core or arivis Cloud for classifying regions. Use it for labeling image regions by predefined classes, using random forest or deep learning. For semantic segmentation models, you need an installation of the 3rd party Python Tools.
Background Subtraction		Uses a rolling ball algorithm to subtract uneven background, then applies a global threshold to segment objects. Use it for images with varying background intensity.
Variance-Based Thresholding		Segments based on intensity changes. Use it for brightfield images where objects are distinguished by variance rather than consistent intensity.
Dynamic Thresholding		Applies local thresholds to different regions to handle inhomogeneous backgrounds. Use it for images with uneven illumination.
Don't show this automatically		Activated: The dialog is not opened automatically anymore for all users and can only be opened with the Explore button. Deactivated: The dialog opens automatically when entering the respective step of the image analysis.
OK		Uses the selected segmentation method for the class and closes the dialog.