End-to-end learning for simultaneously generating decision map and multi-focus image fusion result

The general aim of multi-focus image fusion is to gather focused regions different images generate a unique all-in-focus fused image. Deep learning based methods become the mainstream by virtue its powerful feature representation ability. However, most existing deep structures failed balance quality and end-to-end implementation convenience. End-to-end decoder design often leads unrealistic result because non-linear mapping mechanism. On other hand, generating an intermediate decision map achieves better for image, but relies on rectification with empirical post-processing parameter choices. In this work, handle requirements both output comprehensive simplicity structure implementation, we propose cascade network simultaneously training procedure. It avoids dependence in inference stage. To improve quality, introduce gradient aware loss function preserve information addition, calibration strategy decrease time consumption application multiple fusion. Extensive experiments are conducted compare 19 state-of-the-art 6 assessment metrics. results prove that our designed can generally ameliorate while efficiency increases over 30%