SGNN to Image Fusion Based on Multi-feature Clustering

An image fusion algorithm based on Self-Generating Neural Networks¿SGNN¿ is presented in this paper. Three features are defined and proved to indicate the clarity of an image block. These features are extracted and fed into the neural networks, which learn to produce a Self-Generating Neural Tree (SGNT) to determine the clustering result. registered source images will create SGNTs. As for one of the SGNTs, the subtree with the largest weight of the SGNT includes the clearest image blocks of the correspondent source image. Two-step fusion is proposed, the primary fusion combines the clearest blocks of source images. Then, in secondary fusion, complete the primary fusion image using a weighted average algorithm for source images. Comparing the algorithm proposed in this paper to the Laplacian pyramid and DWT-based ones, experimental results show that the performance of the algorithm proposed in this paper is superior to those two.