AI-ordlista
Den kompletta ordlistan över AI
Interactive Segmentation
Image segmentation process where a user guides the algorithm by providing cues, such as clicks or strokes, to refine the results in real time.
Positive and Negative Clicks
Annotation method where the user clicks on pixels belonging to the object of interest (positive) and on those of the background (negative) to constrain the segmentation model.
Scribbles
Annotations in the form of rough strokes drawn by the user on an image to indicate foreground and background regions, serving as constraints for the algorithm.
Seed Propagation
Technique where starting points (seeds) provided by the user are used to propagate a segmentation label to neighboring pixels based on similarity criteria.
GrabCut
Interactive segmentation algorithm based on graph cuts that uses a Gaussian mixture model to model the foreground and background colors.
Region Growing Segmentation
Iterative method that starts from one or more starting points (seeds) and adds neighboring pixels that satisfy a homogeneity criterion, often defined by the user.
Interactive Watershed
Application of the watershed algorithm where the markers, defining the catchment basins, are placed manually by the user to control over-segmentation.
Bounding Box Refinement
Technique where the user draws a bounding box around the object, and the algorithm refines the segmentation inside this box, often used as a quick first step.
Deep Extreme Cut (DEXTR)
Interactive segmentation method based on deep learning that uses four extreme points (top, bottom, left, right) of the object to generate an accurate segmentation.
Foundation of Active Learning
Principle where the algorithm identifies areas of uncertainty and requests user intervention only for these regions, thus optimizing the annotation effort.
Mask Correction
Iterative process where the user can add or remove regions from an initial segmentation mask, often via clicks, to improve the accuracy of the final result.
Example-based Segmentation
Approach where the user provides examples of pixels belonging to the target class, and the algorithm segments the rest of the image by finding similar pixels based on features like color or texture.
User-Weighted Cost Function
In optimization methods, a cost function where terms related to user constraints are given a high weight, forcing the solution to strictly adhere to the provided indications.
Interactive Multi-Scale Resolution
Strategy that performs an initial segmentation on a low-resolution version of the image for speed, then refines the results at higher resolutions using the same user constraints.
Appearance Model
Statistical model (e.g., Gaussian Mixture Model, histogram) of the color or texture features of a region, built from user annotations to guide the segmentation.
Boundary Editing
Specialized interactive tools that allow the user to directly manipulate the contours of the segmentation mask, for example by pulling/pushing the boundary or adding control points.