Image segmentation is the process of partitioning an image into multiple segments (sets of pixels), typically to locate objects or boundaries. The result is often either: semantic segmentation – every pixel is assigned a class label (e.g., car, road, person, sky), thus segmenting the image into meaningful regions; or instance segmentation – which is like object detection at the pixel level, each object instance gets its own separate mask. There’s also panoptic segmentation which combines the two (all pixels labeled, with instances distinguished for “thing” classes). Traditional approaches used clustering in color space, edge detection + region merging, or graphical models (Markov Random Fields). Modern approaches are dominated by deep learning (Fully Convolutional Networks, U-Net, Mask R-CNN, etc.), where a convolutional network outputs a pixel-wise map of class probabilities. Segmentation is crucial in medical imaging, autonomous vehicles (drivable area vs obstacles), scene understanding, etc.