ISA An algorithm for image segmentation using ants

On one hand, there is the problem to solve which is the image segmentation. It is a low-level processing task which consists in partitioning an image into homogeneous regions. Segmentation can be seen as a combinatorial optimization problem. In fact, considering the huge amount of information that an image carries, it is impossible to find the best segmentation. On the other hand, the reduced individual properties of ants as well as the simplicity of their behaviours led to the design of several methods of optimization by ants such as: optimization by colony of ants (ant colony optimization) and classification by co-operating ants (AntClass, AntTree, AntClust...). In this paper, we present an algorithm for the resolution of the segmentation problem. This algorithm, named ISA (image segmentation using ants), is based on the behavior of ants while cleaning their nest. The image to be segmented represents the environment of the ants. Initially, ants are generated and are positioned randomly on the image. Then they start moving. During its move (searching for a label to the current pixel) an ant can perform a number of tasks such as labeling a pixel, or improving the homogeneity of regions by correcting possible errors (changing the label of a pixel).