FuzzySegNet: Deep Fuzzy Neural Network for Corrupted Image Segmentation

The accurate segmentation of corrupted images is a critical task in the field of computer vision, demanding sophisticated solutions for precise pixel-level analysis. This paper introduces FuzzySegNet, an advanced deep learning architecture specifically designed to address the challenges of corrupted image segmentation. First, FuzzySegNet integrates a robust encoder guided by masked image modeling. This encoder leverages sparse fuzzy logic to extract resilient features from both detailed and contextual information within corrupted images. Second, a robust matrix decoder with a matrix factorization head module is designed to utilize low-rank matrix factorization for robust global feature reorganization. Third, comprehensive evaluations on diverse datasets and real-world corrupted images establish FuzzySegNet as a state-of-the-art framework, showcasing its superiority in handling corrupted image segmentation with remarkable precision. This work not only sets a new standard for corrupted image segmentation but also lays the groundwork for future advancements in the field.