Optimizing Feature Representation via A Nested Network for Object Segmentation
Abdalrahman Alblwi, Kenneth E. Barner · 2022
Automatic object segmentation based on artificial neural networks is a critical task in an array of real-world applications. Localizing and region segmentation is of particular interest, although typical approaches rely on complex networks and/or human interactions. Therefore, various complex networks suffer from suboptimal segmentation due to inaccurate feature extraction. This paper introduces a Multi-Gated Nested Network (MGN-net) that provides precise segmentation performance by capturing relevant contextual information via a channel gating mechanism. Results utilize challenging biomedical image databases, featuring MRI Brain and Chest X-ray images, are presented. The results show that the MGN-net approach subjectively and objectively performs favorably compared to multiple state-of-the-art methods, such as the U2-net and U-net networks.