Research On Image Fusion Algorithms Using Bidimensional Empirical Mode Decomposition

With the explosive growth of complex multi-source image data,multi-view information fleetly and accurately received by image fusion has become the hot research.However,commonly used image fusion algorithm,which is based on wavelet transform or higher level wavelet transform,has the difficulties of wavelet base selection,short of partial correlation for decomposition coefficients and so on,and lowers image fusion effect.As an adaptive multiscale and multi-resolution image analysis technology,Bidimensional Empirical Mode Decomposition(BEMD),which has perspective characteristics of special "biological section" and fitting analysis for nonlinear and nonstationary bidimensional image signal,can receive decomposition coefficients with strong partial correlation.BEMD can efficiently describe and process image characteristics information to obtain a more resultful fusion image.Therefore,introducing BEMD on image fusion area has the broad prospects.Research of image fusion theory,method and application based on BEMD has great significance and value.In this paper,image fusion algorithm based on BEMD has been proposed,in which image fusion rules based on source image characteristics has been deeply studied.The details in this dissertation are as follows:According to pseudo-gibbs phenomena in multispectral image and panchromatic image fusion resulting from wavelet transform,a multi-spectral image and panchromatic image fusion algorithm,which is based on BEMD and Hue,Intensity,Saturation(HIS)transform,is proposed.Firstly multispectral image is decomposed by HIS transform,the received new I component and panchromatic image are respectively resolved by BEMD decomposition to get the multi-scale spectrums of strong local correlation,high frequency and low frequency spectrum coefficients are determined by T test.Then arithmetic average fusion rules is used on low frequency coefficients and spatial frequency fusion rules is applied on high frequency coefficients,the new 'I components are obtained by inverse BEMD transformation on received low and high frequency coefficients,the fused image is obtained by inverse HIS transformation.Results show that the algorithm not only enhance spatial detail information of fusion image,but also keep more spectrum information of source image.In addition,it also can avoid pseudo phenomena caused by traditional wavelet fusion algorithm.1.In allusion to shortcomings of multi-focus image fusion based on wavelet transform and higher level wavelet transform,which has selection difficulties of wavelet base,poor local correlation of image decomposition coefficient,complicated time-consuming decomposition component process,and so on,a multi-focus image fusion algorithm based on BEMD and improve local energy is proposed.First,multi-focus image decomposed by BEMD became into multi-scale spectrum of more local correlation,and then plus or minus phase information of based on the regional spectrum coefficient center pixel,local energy is computed by weighted template of strong correlation,and the maximum standard and weighted average are used to design fusion coefficient selection rules.In the end,the results show the superior image fusion effect than traditional maximum rule,weighted average and wavelet fusion rules.2.Considering that large difference of Synthetic Aperture Radar(SAR)and multi-spectral image source characteristics,serious information loss of fusion results,spectral distortion,low contrast,and so on,the self-adaptive SAR and multi-spectral image fusion algorithm,which is based on BEMD and simulated annealing algorithm,is presented.SAR and multi-spectral image source are respectively resolved by BEMD decomposition to get the multi-scale spectrums of strong local correlation,high frequency and low frequency spectrum coefficients are determined by T test.Then self-adaptive weighted information fusion criterion is applied on low frequency coefficients,more reserved regional correlation information brings more suitable brightness and clarity accord with human visual characteristic in image fusion.Local correlation matching degree of high frequency coefficients are computed,and correlation matching degree threshold is set.The correlation between strong weighted template is used to calculate the local energy,Fusion regulations of high frequency coefficients in different threshold are designed to be selection and weighted rules by regional energy,which is computed by strong regional correlation weighted template.In order to promote effectiveness of threshold selection,simulated annealing algorithm is used to search regional matching threshold.Experimental results show that the algorithm not only effectively overcomes edge distortion and spectral distortions of fusion results in wavelet image fusion algorithm,but also avoids blindness problem of correlation threshold selection for common regional fusion rules.3.In order to avoid information hiding and dark region in multi-scale spectrum caused by BEMD,and further enhance algorithm for real time,and strengthen important information screening accuracy by multi-spectral image and panchromatic image fusion rules,we propose the multi-spectral image and panchromatic image fusion algorithm based on Bidimensional Window Empirical Mode Decomposition(BWEMD)and a particle swarm algorithm.Firstly multispectral image is decomposed by HIS transform,the received new I component and panchromatic image are respectively resolved by BWEMD decomposition to get the multi-scale spectrums of strong local correlation,high frequency and low frequency spectrum coefficients are determined by T test.Then,because of approximate information,fusion rule of low frequency coefficient is designed to be weighted average.Regional correlation matching degree of high frequency and high frequency characteristics,which is calculated by first order Gaussian,are used to design selection and weighted fusion rules.Different threshold of high frequency coefficients is selected by particle swarm optimization algorithm.In the end,new 'I component is obtained by inverse BWEMD transformation on fusion coefficient.In summary,this paper applies multi-scale multi-resolution BEMD decomposition algorithm on multi-source image fusion,high and low frequency coefficients are determined by T test,optimization fusion rules are designed by advantages of BEMD decomposition and multi-source image fusion demand,visual effect and objective evaluation of image fusion is obviously promoted.Our research provides a novel perspective for the traditional image fusion.