Camouflaged Object Segmentation in Images

Date of Award

2019

Degree Name

Master of Computer Science (M.C.S.)

Department

Department of Computer Science

Advisor/Chair

Advisor: Tam Nguyen

Abstract

Camouflaged objects are generally difficult to be detected in their natural environment even for human beings. In this thesis, we propose a novel bio-inspired network, named the CamoNet, that leverages both instance segmentation and adversarial attack for the camouflaged object segmentation. Differently from existing networks for segmentation, our proposed network possesses two segmentation streams: the main stream and the adversarial stream corresponding with the original image and its flipped image, respectively. The output from the adversarial stream is then fused into the main stream's result for the final camouflage map to boost up the segmentation accuracy. We also introduce the Data Augmentation in the Wild to solve the data insufficiency for network training. Extensive experiments conducted on the public CAMO dataset demonstrate the effectiveness of our proposed network. Our proposed method achieves 89% in accuracy, significantly outperforming the state-of-the-arts.

Keywords

Computer Science, Camouflaged Object Segmentation

Rights Statement

Copyright © 2019, author

Link to Full Text

Share

COinS