Research On Multi-focus Image Fusion Based On Depth Neural Network

Multi-focus image fusion is the fusion of clear areas focusing on different targets in the same scene to obtain full-clear images.It is widely used in computer vision,medicine,military and other fields.The technology involves image segmentation,feature extraction and feature level information fusion,among which feature information fusion is the core.However,in order to solve the problem of model design independence and limited application scenarios,the current popular deep neural network technology is introduced.The research contents of this paper include the following aspects:(1)The operation flow of different levels of image fusion methods is analyzed and analyzed.The fusion algorithm is divided into two categories based on transform domain and spatial domain.The processing steps of the two algorithms are introduced as examples.The filters and fusion rules in each algorithm framework are discussed.The design also throws out the human intervention in the algorithm and the failure to reach the universal principle.(2)Aiming at the problem of artificially designing local filters in traditional image fusion methods,an image fusion algorithm based on deep convolutional neural network is proposed.Focus detection is classified as a two-category problem of images.The introduction of neural networks aims to learn the mapping relationship between source images and weight maps.A neural network is designed with reference to the Siamese network structure to train high-quality image blocks and their fuzzy versions,and to encode the above mappings.The experimental results show that the convergence speed of the network model is improved,and the method can provide better fusion performance.(3)The wavelet packet decomposition is combined with the convolutional neural network to solve the problem that the picture after the detail extraction by the convolutional network cannot distinguish the bright area bright spots and the clear area dark spots.A brightness adaptive fusion method based on multi-band image energy similarity metric is proposed.The method is based on image sub-band adaptive selection fusion method based on image scene decomposition,thus avoiding spot and darkness in neural network weight map.The effect of points on the quality of the fused image.The results of multiple experiments show that the method has improved the quality of the results and has been expanded in the application of the scene.