Multi‐focus image fusion through DCNN and ELM

Abstract

The purpose of researches on multi-focus image fusion is to obtain a composed image where the objects are all captured in focus. Compared with the source images, the new one is of richer information and much better visual performance. Deep convolutional neural network (DCNN) and extreme learning machine (ELM) are combined to be a novel model (DCELM) to deal with the issue of multi-focus image fusion. First, the source images are input into DCELM. Then, ELM is responsible for generating random weights between adjacent layers. Moreover, a convolution layer followed by a pooling one forms the basic unit of DCELM, which is used to get the feature maps of the source images from different perspectives. Finally, the above features are classified via ELM, and the information in focus from the source images can be fused into the final fused image. Experimental results demonstrate that the proposed fusion method well combines the better feature extraction ability of DCNN and much faster training speed of ELM, and its performance is superior to current state-of-the-art typical ones.