Research On Multi-focus Images Fusion Algorithm Based On Image Adaptive Decomposition And Multi-directional Features

When shooting with a camera,due to the limitation of the focus range of the lens,it is impossible to obtain a clear image of all objects in the scene.Image fusion technology combines multi-focus images to obtain clear images in the scene,which is convenient for computer processing in the later stage of the image.The multi-focus images fusion algorithm based on image adaptive decomposition and multidirectional features is discussed.The research content of the thesis is as follows:A multi-focus images fusion algorithm based on image adaptive decomposition and multidirectional gradient is proposed.First,an image adaptive decomposition method based on the Gabor directional filter is designed,and filters are designed and preset into the deconvolutional neural network model.When inputting multi-focus images into the network for training,the high-frequency and low-frequency feature maps of multi-focus images are obtained.In the fusion rule design of high-frequency feature maps,because the highfrequency feature maps processed by the directional filter only retain feature information in a specific direction.Therefore,for feature maps in different directions,regional gradient features in different directions are used,and the larger is used to achieve the fusion of highfrequency feature maps.The rule of taking larger regional energy is adopted by lowfrequency feature maps to realize their fusion.The fusion algorithm based on image adaptive decomposition and multi-directional gradient is compared with the existing fusion algorithm based on neural network and directional filtering,the results show that this algorithm is better in terms of information entropy and average gradient,indicating that the image adaptive decomposition method and fusion rules designed by this algorithm are better.A multi-focus images fusion algorithm based on image adaptive decomposition and multi-directional energy is proposed.First,an image adaptive decomposition method based on the Gabor directional filter is designed,and filters are designed and preset into the deconvolutional neural network model.When inputting multi-focus images into the network for training,the high-frequency and low-frequency feature maps of multi-focus images are obtained.In the fusion rule design of high-frequency feature maps,because the highfrequency feature maps processed by the directional filter only retain feature information in a specific direction.Therefore,for feature maps in different directions,the Sum-ModifiedLaplacian in different directions are used,and the larger is used to construct a fusion decision map.The rule of taking a large regional standard deviation is adopted by low-frequency feature maps to realize their fusion.The fusion algorithm based on image adaptive decomposition and multi-directional energy is compared with the existing fusion algorithm based on neural network and directional filtering,the results show that this algorithm is better in terms of information entropy and average gradient,indicating that the image adaptive decomposition method and fusion rules designed by this algorithm are better.A multi-focus images fusion algorithm based on parameter adaptive pulse coupled neural network is proposed.First,an image adaptive decomposition method based on the Gabor directional filter is designed,and filters are designed and preset into the deconvolutional neural network model.When inputting multi-focus images into the network for training,the high-frequency and low-frequency feature maps of multi-focus images are obtained.In the fusion rule design of high-frequency feature maps,because the highfrequency feature maps processed by the directional filter only retain feature information in a specific direction.Therefore,for feature maps in different directions,spatial frequency features in different directions are used to excite the adaptive pulse coupled neural network to achieve the fusion of high-frequency feature maps.Low-frequency feature maps are also fused with an adaptive pulse coupled neural network,and the input excitation is the energy value of the image area.The fusion algorithm based on parameter adaptive pulse coupled neural network is compared with the existing fusion algorithm based on neural network and directional filtering,the results show that this algorithm is better in terms of information entropy and average gradient,indicating that the image adaptive decomposition method and fusion rules designed by this algorithm are better.The fusion algorithm proposed in this thesis can obtain higher quality multi-focus fusion images,which is beneficial to improve the performance of subsequent image processing and recognition.