Comparing Supervised vs. Unsupervised Image Segmentation Methods

This project compares the supervised logistic regression segmentation algorithm against the unsupervised k-means clustering segmentation. We observed that the difference between either method is not very significant. When performed on the 100 test cases for BSD300, the supervised method on average achieved a precision rate of 0.47 and the unsupervised method achieved a precision rate of 0.41. The variance in performance for both method is quite high, indicating that the models performed well for some images but not for other. This effect seems more prevalent for the supervised result. The standard deviation values for the precision rate are σsupervised = 0.18, σunsupervised = 0.16