Generated Pseudo-Labels Guided by Background Skeletons for Overcoming Under-Segmentation in Overlapping Particle Objects.
Jinshi LiuZhaohui JiangTing CaoZhiwen ChenChaobo ZhangWeihua GuiPublished in: IEEE Trans. Circuits Syst. Video Technol. (2023)
Keyphrases
- background pixels
- bounding box
- object segmentation
- textured objects
- complex background
- multiple objects
- object regions
- object contours
- figure ground segmentation
- image labeling
- foreground objects
- segmentation algorithm
- segmented images
- distance maps
- segmentation method
- image regions
- foreground and background
- cluttered background
- d objects
- spatially correlated
- outdoor scenes
- training data
- level set
- input image
- low depth of field
- foreground background separation
- image segmentation
- skeleton extraction
- training set
- moving objects
- background modeling
- motion segmentation
- labeling problems
- medical images
- pixel level
- semantic segmentation
- object region
- object shapes
- multiscale
- shape prior
- label field
- dynamic scenes
- superpixels