Abstract:The existence of camouflage targets is widespread in the natural world, as they blend seamlessly or closely resemble their surrounding environment, making it difficult for the human eye to identify them accurately. In camouflage target segmentation, challenges often arise from the high similarity between the foreground and background, resulting in segmentation errors, imprecise edge detection, and overlooking of small targets. To address these issues, this paper presents a robust localization-guided dual-branch network for the recognition of camouflaged targets. Two crucial branches, i.e., a localization branch and an overall refinement branch are designed and incorporated. The localization branch achieves accurate preliminary localization of camouflaged targets by incorporating the robust localization module, which integrates different high-level feature maps in a partially decoded manner. The overall refinement branch optimizes segmentation accuracy based on the output predictions of the localization branch. Within this branch, the edge refinement module is devised to effectively reduce false negative and false positive interference. By conducting context exploration on each feature layer from top to bottom, this module further enhances the precision of target edge segmentation. Additionally, our network employs five jointly trained output prediction maps and introduces attention-guided heads for diverse prediction maps in the overall refinement branch. This design adjusts the spatial positions and channel weights of different prediction maps, generating output prediction maps based on the emphasis of each output, thereby further strengthening the perception and feature representation capabilities of the model. To improve its ability to generate highly confident and accurate prediction candidate regions, tailored loss functions are designed to cater to the objectives of different prediction maps. We conducted experiments on three publicly available datasets for camouflaged object detection to assess our methodology and compared it with state-of-the-art network models. On the largest dataset COD10K, our method achieved a Structure-measure of 0.827 and demonstrated superior performance in other evaluation metrics, outperforming recent network models.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the accurate segmentation of camouflaged targets in natural environments. Due to the high similarity between camouflaged targets and the background, traditional methods face many challenges in identifying these targets, such as blurred edges, complex and changeable backgrounds, and small targets being easily overlooked. To address these issues, the paper proposes a Robust Localization - Guided Dual - Branch Network (RLGDBNet), aiming to simulate the process of human observing camouflaged targets, achieve the initial localization and detailed feature distinction of camouflaged targets, and thus improve the segmentation accuracy and detection effect. Specifically, the network design in the paper contains two key branches: 1. **Localization Branch**: This branch realizes the initial localization of camouflaged targets by introducing the Robust Localization Module (RLM) and using the partial decoding and fusion of different high - level feature maps. This module can capture the position and boundary information of the target and generate an initial localization prediction map to provide guidance for the subsequent refinement branch. 2. **Overall Refinement Branch**: Based on the output of the localization branch, this branch further optimizes the segmentation result. In this branch, an Edge Refinement Module (ERM) is designed to gradually explore the context information from the top layer to the bottom layer, reduce false detections and missed detections, and enhance the segmentation accuracy of the target edge. In addition, an Attention - Guided Head (AG - Head) is introduced to generate a more accurate output prediction map by adjusting the spatial positions and channel weights of different prediction maps. Through the design of these two branches, the paper aims to overcome the limitations of traditional methods in handling the camouflaged target segmentation task, especially in complex backgrounds. The experimental results show that this method has achieved excellent performance on multiple public datasets. Especially on the largest dataset COD10K, it has achieved a Structure - measure of 0.827, and also outperforms recent network models in other evaluation indicators. In conclusion, through the innovative network architecture and module design, this paper significantly improves the accuracy and robustness of camouflaged target segmentation, providing new ideas and technical support for research and applications in related fields.

Robust Localization-Guided Dual-Branch Network for Camouflaged Object Segmentation

Detecting Camouflaged Objects via Multi-Stage Coarse-to-Fine Refinement

A bioinspired three-stage model for camouflaged object detection

Double-Branch Camouflaged Object Detection Method Based on Intra-Layer and Inter-Layer Information Integration

Towards Deeper Understanding of Camouflaged Object Detection

Attention guided multi-level feature aggregation network for camouflaged object detection

Efficient Camouflaged Object Detection Network Based on Global Localization Perception and Local Guidance Refinement

Edge-Aware Mirror Network for Camouflaged Object Detection

Dual Guidance Enhancing Camouflaged Object Detection Via Focusing Boundary and Localization Representation

CAMOUFLAGE-Net: comprehensive advanced model for optimal camouflaged target detection and analysis using groundbreaking elements

Detecting camouflaged objects via cross-level context supplement

Camouflaged Instance Segmentation Via Explicit De-Camouflaging

Edge-Guided Camouflaged Object Detection Via Multi-Level Feature Integration.

FRBNet: feature-iterative reinforcement and boundary-directed network for camouflaged object detection

SPANet: Spatial perceptual activation network for camouflaged object detection

CFANet: A Cross-layer Feature Aggregation Network for Camouflaged Object Detection

Adaptive Query Selection for Camouflaged Instance Segmentation

GLCONet: Learning Multi-source Perception Representation for Camouflaged Object Detection

Local to global purification strategy to realize collaborative camouflaged object detection

Camouflaged Object Detection via Dual-branch Fusion and Dual Self-similarity constraints

Lightweight camouflaged object detection model based on multilevel feature fusion