Research On Image Fusion Method Based On Bidimensional Empirical Mode Decomposition

Image fusion can be defined as the process by which two or more images are combined to produce a new image that integrates complementary and redundant information from the sources.Image fusion has rich application sceneries,such as military domain and civil domain,so the study of image fusion technology has important significance.Among all the image fusion techniques,the method based on the multi-dimension analysis has been the research focus in the field in recent years.The key point of image fusion scheme based on the multi-resolution analysis is the selection of decomposition tools and the decision of fusion rules.Bidimensional empirical mode decomposition(BEMD)is an adaptive data decomposition representation,as it don't need predetermined filters or wavelet functions,and has better performance for nonlinear and non-stationary time series analysis than conventional wavelet transform.This paper conduct a systematic study of image fusion method in BEMD domain,the work is summarized as follows:Firstly,the existing BEMD decomposition tools are divided into four types:line and column crossed-used EMD(LCEMD),complex empirical mode decomposition(CEMD),interpolation function based EMD(IFEMD)and local average value based EMD(LAVEMD)according to its application sceneries in image fusion.The decomposition characteristics of each decomposition tool have been summarized in detail.And the specific instances are used to demonstrate the difference of each decomposition tool obviously.Secondly,a novel image fusion method based on local regional feature and improved bidimensional empirical mode decomposition is proposed.In order to avoid the mismatching IMFs,either in number or their frequency,which is produced by IFEMD,a new solution to this problem is proposed using a fixed number of iterations and coordinated operation in sifting process.Experimental results show that the proposed algorithm has good performance for multi-focus image fusion,remote sensing image fusion and medical image fusion.It can fully get the complementary features and keep effective information of source images.Thirdly,a new medical image fusion method based on the characteristics of local average value based empirical mode decomposition(LAVEMD)and the properties of pulse coupled neural network(PCNN)is proposed.The LAVEMD can separate the image high frequency detail information and low frequency detail information.For high frequency information,the self-adaptive PCNN model is used to choose the clear part of the image.And the method based on local energy is adopted to enhance the details for low frequency information.Experimental results show that the proposed algorithm has good performance for multi-modality medical image fusion.It can provide far more comprehensive information and improve reliability of clinical diagnosis and therapy.