Multi-source Image Fusion Based On Pulse Coupled Neural Network And Multi-scale Analysis

Because of the difference of imaging mechanisms and conditions,single sensor image cannot reflect the whole detailed information of scene effectively;therefore,the techniques for the reservation of complementary information and the removal of redundant information are becoming significant and fundamental parts in image processing.Image fusion can effectively address this problem by producing a complete and accurate image whose complementary features information are extracted and fused from the multi-source images;as a result,image fusion technique is widely used in military,medicine,remote sensing,surveillance and other fields.In recent years,the rapidly development of image representation and analysis techniques constantly generates challenges and probabilities for image fusion;besides,the requirement of high quality images also drives image fusion becoming a hot research field.In this thesis,pulse coupled neural network(PCNN)and multi-scale analysis techniques are employed as the basic tools;this work focuses on color image feature extraction and fusion,and detailed or regional feature extraction in image fusion techniques.The researches of this thesis include the neurodynamics analysis of PCNN and its simplified versions,the feature extraction and algorithm optimize for multi-focus image fusion method,the color space selection and color feature extraction for color multi-focus image fusion,the spectral information and detailed feature fusion for remote sensing image,the detailed and regional feature fusion for visible and infrared image.The primary work and contributions of the thesis are summarized as follows:(1)This thesis focuses on the dynamic mechanics of the neurons in PCNN,PCNN and ICM,respectively.Firstly,the firing(activated)conditions of neurons in non-linking and linking states are analyzed according to the three models;then,the rationalities of the firing condition analysis are proved by the firing period analysis of neuron in this thesis.(2)For the characteristics of regional features in multi-focus images,this thesis proposes a block image fusion method based on spatial frequency and S-PCNNwhose parameters are employed by PSO.Besides,the thesis also represents a novel fitness function by the combination of several image quality evaluation indexes,which can improve the efficient of the proposed method and enhance its adaptability and robustness.(3)Based on the analysis of different color spaces,this thesis proposes a method which fuses the three color components by different methods according to different characteristics of the three color components.The proposed method utilizes the image decomposition capacity of NSST and the feature extraction capability of PCNN to propose an image fusion method.The performances of the proposed method in different color space are tested,and feasibility of the reported idea is proved by the experiments in this thesis.(4)For the difference of spectral information and detailed features in different source remote sensing images,this thesis proposes a remote sensing image fusion method based on NSST and PCNN in CIE Lab color space.The images are decomposed by NSST to get low and high frequency sub-images,and the former is fused by ICM,and the latter is fused by PCNN.The experiments show that the proposed method can fuse the spectral information and detailed features of remote sensing images.(5)Infrared image has saliently regional features and visual image has abundantly detailed information;for the mentioned characteristics,this thesis proposes a method which firstly uses SWT to decompose the source image into sub-images;and then DCT and LSF are employed to extract the regional features of the source images;at last,the image coefficients are fused by fusion rules.The experiments show that the proposed method can extract and fuse the regional and detailed features of the source images.The research of this thesis focus on the mechanics of PCNN models,the color features extrication of color images,and the detailed and regional features extrication of images,which can be summarized as three aspects: the dynamic mechanics of the neurons in PCNN,PCNN and ICM are analyzed,which will help to understand the operating mechanism of these models and can provide the theoretical foundation fortheir applications;the influence of different color spaces for image fusion are analyzed,and CIE Lab also is applied into color image fusion methods in this thesis;the feature extraction of image fusion is studied in this thesis to propose several feasible image fusion methods.