An interactive ground truth labelling program that allows a user to mark bumps and holes on a set of Braille images. The labeled regions are then compared against pre-processed algorithm data. This then determines how well the algorithm has performed compared to what the user has determined as the ground truth.
This project utilizes the OpenCV library. It is a library made for real-time computer vision functionality and machine learning analysis.
F-Measure is the quantitative measure of how well a given machine learning algorithm has performed based on how correct and incorrect predictions.
See definitions for TP, FN and FP in the Confusion Matrix section below.
Recall = TP / (TP + FN)
Precision = TP / (TP + FP)
Fβ = ( 1 + β 2 ) + ( Precision * Recall ) / ( β 2 * Precision) + Recall ), where β = 1
The value of Fβ is the percentage accuracy of the machine learning algorithm.
A confusion matrix let's you visualize the performance in a numerical table format.
| Legend | |
|---|---|
| TP | True positive, the algorithm has detected what it is intended to detect. |
| FP | False positive, the algorithm has detected what it is looking for. But it is not actually there. |
| FN | False negative, the algorithm was supposed to detect the target, but instead did not. |
| Truly a bump | Truly a hole | Truly None | |
|---|---|---|---|
| Detected as a bump | TP | FP | FP |
| Detected as a hole | FP | TP | FP |
| Detected as null | FN | FN | Null |
