A Multi-step Fusion Network Based on Environmental Knowledge Graph for Camouflaged Object Detection
Zheng Wang (Tianjin University),* Wenjun Huang (Tianjin University), Ruoxun Su (Tianjin University), Xinyu Yan (Tianjin University), Meijun Sun (Tianjin University)
The 34th British Machine Vision Conference
Abstract
Due to the high similarity in color and texture between camouflaged objects and noise backgrounds, existing single-step detection methods often fail especially when the camouflage level of objects is high. However, with prior knowledge of the environment, humans can effectively distinguish camouflaged objects, for example, when humans see snowy ground, they spontaneously associate that white rabbits might be concealed there. In this paper, we propose an Environmental Knowledge-guided Multi-step Network (EKNet) to simulate this mechanism. To extract prior knowledge of the background, we construct a knowledge graph with information extracted from the image and generate a relevance score matrix (RS) for prior knowledge and the camouflaged object with GCN as the correlation scoring matrix generation module (CSM). After that, we fuse the RS with Canny edge-enhanced features, which guides the model to detect camouflaged objects more accurately by observing the background information with edge semantics as the knowledge integration module (KIM). To our knowledge, this work is the first to introduce environmental knowledge to guiding camouflaged object detection (COD). Extensive experiments on three benchmark datasets show that our EKNet outperforms 15 existing state-of-the-art methods under four widely-used evaluation metrics.Video
Citation
@inproceedings{Wang_2023_BMVC,
author = {Zheng Wang and Wenjun Huang and Ruoxun Su and Xinyu Yan and Meijun Sun},
title = {A Multi-step Fusion Network Based on Environmental Knowledge Graph for Camouflaged Object Detection},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year = {2023},
url = {https://papers.bmvc2023.org/0265.pdf}
}
Copyright © 2023
The British Machine Vision Conference is organised by