Document Type : Review Article

Author

• Ali Jelokhani

School of Electrical and Computer Engineering, University of Tehran, Tehran, Iran

Abstract

Camouflaged Object Segmentation (COS) aims to segment objects at the pixel level where objects have minimal visual contrast with their surroundings. Object boundaries are often weak, and background distractors can cause incomplete masks and false positives. Though deep learning techniques have significantly boosted the development of COS, literature is scattered and heterogeneous with different deep learning architectures, training procedures, and benchmark protocols. To address this, we conducted this systematic literature review to synthesize existing literature on Camouflaged Object Segmentation techniques published between 2020 and 2026. We analyzed 38 eligible studies, extracted standardized study characteristics (datasets/protocols, core methodological ideas, architectural choices, and reported improvements), and summarized them in a method-centric taxonomy and an appraisal table. The literature shows that consistent improvements are achieved through boundary-aware and structure-aware refinement, multi-scale/coarse-to-fine reasoning, transformer-based global context and local detail recovery, and uncertainty-aware iterative refinement, including diffusion- and unfolding-based approaches.

Keywords

• camouflaged object segmentation
• systematic literature review
• PRISMA
• segmentation
• multi-scale learning
• transformers
• diffusion models