Image segmentation using deep learning is a crucial research area in computer vision that focuses on partitioning images into meaningful regions or objects at the pixel level. Foundational works introduced fully convolutional networks (FCNs) that replaced traditional fully connected layers with convolutional layers to produce dense predictions. Subsequent research enhanced segmentation accuracy through architectures such as U-Net, SegNet, DeepLab (with atrous convolutions and spatial pyramid pooling), and Mask R-CNN, enabling precise delineation of objects, even in complex scenes. Recent advances incorporate attention mechanisms, transformer-based models, and multi-scale feature fusion to capture long-range dependencies and improve boundary refinement. Applications span medical imaging, autonomous driving, satellite and aerial imagery, video analysis, and augmented reality, where accurate segmentation is critical for downstream tasks such as object recognition, scene understanding, and robotic perception. Current research also explores lightweight and real-time segmentation models suitable for edge deployment, semi-supervised and self-supervised learning to reduce annotation costs, and multi-modal segmentation combining RGB, depth, and infrared data.