In PyTorch, images are represented as four-dimensional tensors with shape [B, C, H, W] (batch, channels, height, width), and image processing operations are essentially mathematical transformations of tensors.