AUTHOR=Ike Chidiebere Somadina , Muhammad Nazeer , Bibi Nargis , Alhazmi Samah , Eoghan Furey TITLE=Discriminative context-aware network for camouflaged object detection JOURNAL=Frontiers in Artificial Intelligence VOLUME=Volume 7 - 2024 YEAR=2024 URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2024.1347898 DOI=10.3389/frai.2024.1347898 ISSN=2624-8212 ABSTRACT=Animals in a variety of environments use camouflage as a flexible adaptation to enable a variety of protection mechanisms, including background matching, disruptive colouration, countershading, and masquerade. These adaptive tactics are intended to confuse would-be observers and make it difficult for them to detect or accurately discern the presence of camouflaged organisms. Camouflage Object Detection (COD) techniques are specifically designed to address this challenge by precisely identifying object regions that seamlessly blend into their natural surroundings. The main target is accurately detecting and segmenting the hidden object prediction map. Most of these techniques fail to meet the benchmark requirements of COD. Specifically, natural systems comprise a variety of noisy inferences that contribute to object concealment. Therefore, finding a signal hidden in extreme situations becomes a challenge. Inspired by this observation, the salient information of the camouflaged image is restored in two distinct stages. First, an adaptive restoration block intelligently learns feature attention weights. These weighting strategies pay more attention to effective information. This is achieved by assigning different weighted information to channel-and pixel-wise features. This contributes to the representative ability of convolutional neural networks and their robustness in dealing with various types of information. Second, a cascaded detection module fortified with an enlarged receptive field block was used to refine the camouflaged object prediction map with clear boundaries. Without any post-processing, the proposed method generates an accurate saliency map with contextual details and precise boundaries. To validate the effectiveness of our proposed discriminative context-aware network (DiCANet), extensive experiments on three challenging COD datasets-the CAMO, CHAMELEON, and COD10K-have shown state-of-the-art deep learning performances.