Efficient structural segmentation of scene image with genetic algorithm for merging boundary regions

摘要

In this paper, an efficient structural segmentation of color scene image with genetic algorithm for merging boundary regions is proposed. The segmentation is divided into four main processes. First, pseudo KL transformation is applied to the original image to reduce strong correlation of brightness between RGB planes. Next, on the transformed planes the four tree-structural splitting is applied to the initial block images, then neighboring regions of the objective region are merged into a region using simple criteria for density values. Finally, the genetic algorithm is used to optimize the segmentation of ambiguous boundary regions. The method is computationally simple compared with the existing methods. The effectiveness of segmentation is demonstrated with several color scene images in the simulations.