Region Quality Based Scale-aware Selection for Hierarchical Image Segmentation
Haonan Zhu, Bo Peng, Hongjun Wang · 2021
In this paper, we propose a scale selection method for hierarchical image segmentation task, which is based on the quality of regions. Firstly, the tree-like representation of hierarchical segmentation result is used to establish the region level relationship. Then the region quality is calculated based on defined region features and a graphical model is constructed to obtain the optimal scale labels for each of the lowest level regions. Lastly, the final segmentation is achieved by combining the corresponding regions from the optimal hierarchy. The experimental results show that the proposed method outperforms the traditional thresholding method for scale selection, and can improve the local and global segmentation quality. As a post-processing method, it can largely improve the output quality of the hierarchical segmentation in vision tasks.