. Author manuscript; available in PMC: 2018 Jan 18.

Published in final edited form as: IEEE Trans Image Process. 2016 Jul 18;25(10):4596–4607. doi: 10.1109/TIP.2016.2592704

TABLE III.

Results on BSDS500 with Gaussian white noise with different variances $σ_{n}^{2}$ of hierarchical merge tree model with iterative segmentation accumulation. The training uses noise-free BSDS300 training images. The segmentation covering (Covering), the probabilistic Rand index (PRI), and the variation of information (VI) are reported for optimal data set scale (ODS) and optimal image scale (OIS).

Covering

PRI

σ_{n}^{2}

Iter.

ODS

OIS

ODS

OIS

ODS

OIS

0.001

0
10

0.457
0.617

0.459
0.665

0.587
0.836

0.587
0.860

1.929
1.576

1.917
1.411

0.01

0
10

0.343
0.574

0.344
0.606

0.394
0.805

0.394
0.815

2.278
1.754

2.271
1.581