How to make a simple threshold labeler using the custom algorithms in image labeler.

Can anyone help me understand the custom algorithms in image labeler? I've looked through the help info and I've looked at the inline notes within the script template that is generated yet I still don't quite understand how to even start making the algorithm. All I would like to do is run my custom algorithm, have it ask for a min max value, and then classify the pixels within that range. E.g. min value -25 max value 25 label pixels "good_data". It is easy to do this outside of the algorithm, however I'm trying to get it to function within image labeler so it will combine these classified pixels with other manually classified pixels. Any help would be appreciated.

Answer 1

[Aishwarya]

Hi Jordan,

As per my understanding, you wish to integrate a simple thresholding algorithm into the Image Labeler App, such that you are being able to input minimum and maximum threshold values, classifying pixels within this range as "good data" using this algorithm.

In this case, for implementing this custom algorithm in the Image Labeler App, here are some suggestions and resources:

1. Label Type: As the algorithm requires each pixel to be labeled, the label type to be considered is “Pixel Label”.

2. Create Custom Algorithm Workflow: Refer to the below MathWorks documentation on how to create automation algorithm for labelling: https://www.mathworks.com/help/vision/ug/create-automation-algorithm-for-labeling.html

3. “vision.labeler.AutomationAlgorithm” class: The “MyCustomAlgorithm” template file uses “vision.labeler.AutomationAlgorithm” class for defining custom automation algorithm. Inside this class, consider following suggestions to implement the required algorithm:

• Use “settingsDialog” function to define a dialog box to ask user input for minimum and maximum threshold of pixels. For more information on the "settingsDialog" function, please refer to the documentation here: https://www.mathworks.com/help/vision/ref/settingsdialog.html
• In “run” function, implement the threshold algorithm for each frame and save corresponding labels in the "autoLabels" variable. You can choose to store the labels as a categorical matrix, structure array, or table. For details on the "run" function, please consult the documentation here: https://www.mathworks.com/help/vision/ref/run.html

4. Further, I recommend you reviewing the example implementation provided in the MathWorks Documentation below, specifically the section titled "Integrate Pixel Segmentation Algorithm Into Ground Truth Labeler". This example demonstrates a workflow for integrating a custom algorithm for image segmentation: https://www.mathworks.com/help/driving/ug/automate-ground-truth-labeling-for-semantic-segmentation.html

Hope this helps!

Regards,

Aishwarya