Ant algorithms for image feature extraction

This paper extends on previous work in applying an ant algorithm to image feature extraction, focusing on edge pattern extraction, as well as the broader study of self-organisation mechanisms in digital image environments. A novel method of distributed adaptive thresholding is introduced to the ant algorithm, which enables automated distributed adaptive thresholding across the swarm. This technique is shown to increase performance of the algorithm, and furthermore, eliminates the requirement for a user set threshold, allowing the algorithm to autonomously adapt an appropriate threshold for a given image, or data set. Additionally this approach is extended to allow for simultaneous multiple-swarm multiple-feature extraction, as well as dynamic adaptation to changing imagery.

Highlights
► We propose swarm algorithms for the extraction of features in an image.
► We study self-organization of elemental processing units.
► Our algorithm implements autonomous adaptation with automatic learning of thresholds for optimal segmentation.
► The proposed method works with a multitude of swarms of particles and changing image characteristics.