A proposed adaptive image segmentation method based on Local Excitatory Global Inhibitory region growing

Abstract

Image segmentation is an indispensable first step in many image processing tasks. Many attempts have been made over the time including traditionally approaches (i.e. threshold-based, edge-based, and region growing) to modern methods of machine learning and neural networks. However, the final solution hasn't be found as yet. Recently, the Local Excitatory Global Inhibitory Oscillator Network (LEGION) has been proposed aimed to solve the problem. The LEGION has been developed for over a decade and has various ways of advancement. The all-digital model, a hybrid of LEGION and region growing, has been done in order to overcome the analog operation of the origin. However, there is an issue still exist in the origin and all of its advancements. It is the fragmentation which results from the incorrect chosen parameters. In this paper, we proposed an adaptive image segmentation method which has dynamic parameters in order to get the best performance. Our approach is based on the digital hybrid of LEGION and region growing, and the parameters are not chosen manually but be computed from the contents of image.