Unsupervised hierarchical image segmentation through fuzzy entropy maximization

作者：

Highlights：

• We present an unsupervised multilevel segmentation scheme for automatically segmenting grayscale and color images.

• Fuzzy 2-partition entropy is combined with Graph Cut to form a bi-level segmentation operator that splits a given region into 2 parts based on both global optimal threshold and local spatial coherence.

• A multilevel segmentation scheme iteratively performs on selected regions and color channels, producing a coarse-to-fine hierarchy of segments.

• The presented algorithm is evaluated using the Berkeley Segmentation Database and achieves competitive results compared with the state-of-the-art methods.

摘要

•We present an unsupervised multilevel segmentation scheme for automatically segmenting grayscale and color images.•Fuzzy 2-partition entropy is combined with Graph Cut to form a bi-level segmentation operator that splits a given region into 2 parts based on both global optimal threshold and local spatial coherence.•A multilevel segmentation scheme iteratively performs on selected regions and color channels, producing a coarse-to-fine hierarchy of segments.•The presented algorithm is evaluated using the Berkeley Segmentation Database and achieves competitive results compared with the state-of-the-art methods.

论文关键词：Image segmentation,Superpixel,Fuzzy partition,Graph cut

论文评审过程：Received 5 February 2016, Revised 5 March 2017, Accepted 6 March 2017, Available online 9 March 2017, Version of Record 30 March 2017.

论文官网地址：https://doi.org/10.1016/j.patcog.2017.03.012