Volume Image Segmentaton by Dual Multi-Scale Morphological Reconstructions

Performing image preprocessing for specific targets, e.g., foreground (FG) segmentation and feature extraction, on the scale of databases is challenging. For volume image FG segmentation, we proposed to utilize dual multi-scale graylevel morphological open/close by reconstruction to simulate background (BG) gray-level variational mesh to identify FG regions. It is developed from a global perspective on image FG/BG gray-levels and is carried out with regular operations. To evaluate FG segmentation performance, the probability of coherent segmentation labeling, normalized probability random index, with ground-truth ones is computed for comparisons. For assessing the segmentation capability in dealing with volume images, content-based image retrieval is carried out for performance evaluations. Experiments showed that the proposed FG segmentation method outperforms previous ones with 21% PRI improvement and also improves the retrieval precision-recall performance up to 31%.