Fuzzy Min-Max Neural Network for Image Segmentation

In this work a new fuzzy min-max neural network for color image segmentation, called FMMIS neural network, is proposed. The FMMIS algorithm uses seed pixels to grow hyperboxes, and a criterion of homogeneity for controlling the size of these hyperboxes. The algorithm has been implemented for 2D images and tested on the segmentation of live and dead knots in images of wood boards. On a test set, all knots were correctly detected and most of them were precisely segmented (the area recognition rate was 91%). The method is very fast and may be applied on real-time visual inspection systems.