Unsupervised methods
Unsupervised methods, on the other hand do not require you to mark the area or choose regions manually. The process of identifying segments is fully automatic, with some hyper-parameters available to set the minimum segment size and detail level.
The graph-based approach
This time, will be using the Felzenszwalbalgorithm, which is an unsupervised and efficient graph-based approach. It was proposed by P.F Felzenszwalb and Huttenlocher in 2004 and has been actively used in computer vision since. The benefits of using the Felzenszwalb algorithm are as follows:
- A small number of hyperparameters
- Fast and linear execution time
- Preserves details in low variability areas
Julia implements Felzenszwalb algorithms in the felzenszwalb function, which is part of the ImageSegmentation package.felzenszwalb has two parameters:
- The threshold for the region merging step; the bigger the value, the larger the segment, which is usually set from 10 to 500
- Minimum (optional) segment size; usually...
