Neural Architectures for Mathematical Morphology and Fuzzy Morphology

In this paper, we first propose a general neural structure along with a set of training algorithms for basic binary morphological operations. The proposed algorithms can train the neurons in only a single iteration and use simple arithmetic computation in the testing phase. This feature overcomes most of other models with multiple training iterations in speeding up neuron operations. Besides, the concept of fuzzy morphology is incorporated into the design of neural architecture. This is important in interpreting binary images with noise and gray-scale images in which uncertainty are embeded. The proof of proposed algorithms are provided and examples of different operations are illustrated.