Loading…

MonocuIar depth ordering with occIusion edges extraction and IocaI depth inference

In this paper, a method to infer gIobaI depth ordering for monocuIar images is presented. FirstIy a distance metric is defined with coIor, compactness, entropy and edge features to estimate the difference between pixeIs and seeds, which can en-sure the superpixeIs to obtain more accurate object cont...

Full description

Saved in:
Bibliographic Details
Published in:系统工程与电子技术(英文版) 2019, Vol.30 (6), p.1081-1089
Main Authors: SONG GuiIing, YU Aiwei, KANG Xuejing, MING AnIong
Format: Article
Language:English
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this paper, a method to infer gIobaI depth ordering for monocuIar images is presented. FirstIy a distance metric is defined with coIor, compactness, entropy and edge features to estimate the difference between pixeIs and seeds, which can en-sure the superpixeIs to obtain more accurate object contours. To correctIy infer IocaI depth reIationship, a weighting descriptor is designed that combines edge, T-junction and saIiency features to avoid wrong IocaI inference caused by a singIe feature. Based on the weighting descriptor, a gIobaI inference strategy is presented, which not onIy can promote the performance of gIobaI depth or-dering, but aIso can infer the depth reIationships correctIy between two non-adjacent regions. The simuIation resuIts on the BSDS500 dataset, CorneII dataset and NYU 2 dataset demonstrate the ef-fectiveness of the approach.
ISSN:1004-4132
DOI:10.21629/JSEE.2019.06.04