Image Segmentation via Fusion Neural Networks

This paper presents a self-organizing fusion neural network (SOFNN) which is effective in performing fast image segmentation. Based on a counteracting learning strategy, SOFNN employs two parameters that together control the learning rate in a counteracting manner to achieve free of over-segmentation and under-segmentation. Regions comprising an object are identified and merged in a self-organizing way, and the training process will be terminated without manual intervention. Because most training parameters are data-dependent, implementation of SOFNN is simple. Unlike existing methods that sequentially merge regions, all regions in SOFNN can be processed in parallel fashion, thus providing great potentiality for a fully parallel hardware implementation. In addition, not only the immediate neighbors are used to calculate merging criterion, but the neighboring regions surrounding the immediate regions are also referred. Such extension in adjacency helps achieve more accurate segmentation results.