slic

SLIC, short for Simple Linear Iterative Clustering, is a method for generating an over-segmentation of an image into perceptually meaningful regions called superpixels. It clusters pixels in a five-dimensional space that combines color information (typically in a perceptually uniform color space such as CIELAB) with pixel coordinates, using an iterative, k-means–like procedure. The goal is to produce a fixed number of roughly uniform-sized, compact superpixels that align well with image boundaries.

The standard algorithm operates on a grid of initial center seeds spaced approximately S = sqrt(A/K) pixels apart, where A is the image area and K is the desired number of superpixels. Centers are perturbed to lie in low-gradient regions. The clustering is performed in a local window around each center to reduce computation. Each pixel is assigned to the nearest center by a distance measure that combines color similarity and spatial proximity. A compactness parameter m controls the trade-off between color similarity and spatial proximity; larger m yields more spatially compact superpixels. After assignment, cluster centers are updated as the mean color and position of their members. The process is iterated a fixed number of times (typically 10) or until convergence.

SLIC is valued for its speed and ability to produce nearly uniform, edge-adhering superpixels with a simple implementation. Variants such as SLICO, which operates with no fixed compactness parameter, adapt the balance automatically. SLIC is widely implemented in image processing libraries and is commonly used as a preprocessing step for segmentation, object recognition, and tracking. Limitations include sensitivity to the chosen color space, difficulties with very textured regions, and potential leakage across weak image gradients.