Training

Random Forest Classifier

The Random Forest Classifier is a meta estimator that fits a number of decision tree classifiers on various sub-samples of the dataset and uses averaging to improve the predictive accuracy and control over-fitting. The implementation follows closely the implementation of sklearn

Currently there are two modes of Training for the Random Forest Classifier: A fast one with medium accuracy and a slow one with higher accuracy. The maximum depths of the tree are 10 and 17 respectively.

The Training part doesn’t yeld results. You need to execute the prediction after training to see results.

Hough Circle Transformation

Hough Circle Transformation can be used to detect circles in images. In this training config only one Slice of an image can be chosen. After a successful Hough Transformation the image with its annotations is getting displayed.

This step is for trying on only one image before using on a whole stack of slices.

A tutorial about the implementation of the Hough Circle Transformation can be found here

Note

Hough Circle Transformation is an operation that does not require training. It is currently listed under “Training” aswell as under “Prediction”. During “Training” the Transformation is calculated and the result is shown. “Prediction” only contains a download option for the results.

Ellipse Detection

Ellipse Detection can be used to detect circular and elliptical shapes in images.

The calculated image annotations can be downloaded.

This follows the approach by OpenCV

Note

This does not require training. This is a classic detection method. First a Canny filter is deployed. On this image all contours are getting found with the algorithm from Suzuki et.al. called Topological structural analysis Then it fits an ellipse around a set of 2D points. The function calculates the ellipse that fits a set of 2D points created from the contour.

Example of detecting ellipses with above parameters from an image.