search for




 

Semantic Segmentation Using Trade-Off and Internal Ensemble
Int. J. Fuzzy Log. Intell. Syst. 2018;18(3):196-203
Published online September 25, 2018
© 2018 Korean Institute of Intelligent Systems.

Wang-Su Jeon1, Grzegorz Cielniak2, and Sang-Yong Rhee3

1Department of IT Convergence Engineering, Kyungnam University, Changwon, Korea
2School Computer Science, University of Lincoln, Lincoln, UK
3Department of Computer Engineering, Kyungnam University, Changwon, Korea
Correspondence to: Sang-Yong Rhee
(syrhee@kyungnam.ac.kr)
Received July 21, 2018; Revised September 6, 2018; Accepted September 18, 2018.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
The computer vision consists of image classification, image segmentation, object detection, and tracking, etc. Among them, image segmentation is the most basic technique of the computer vision, which divides an image into foreground and background. This paper proposes an ensemble model using a concept of physical perception for image segmentation. Practically two connected models, the DeepLab and a modified VGG model, get feedback each other in the training process. On inference processing, we combine the results of two parallel models and execute an atrous spatial pyramid pooling (ASPP) and post-processing by using conditional random field (CRF). The proposed model shows better performance than the DeepLab in local area and about 1% improvement on average on comparison of pixel-by-pixel.
Keywords : Convolution neural network, Correlation, Internal ensembles semantic segmentation, Conditional random field