Algorithm Research Of Image Fusion Based On Transform Domain

Multi-source image fusion technique is to generate a fused image with abundantinformation from images with same scene obtained from different sensors, it can reduceredundancy, improve the round-the-clock working capability of systems and improve theability of target recognition. This technique is wide used in military fields, medical imaging,intelligent transportation system, safety supervision system and so on, it plays an essentialpart in national defense and economic construction.This thesis first makes notes on the definition, research background and researchsignificance of multi-source image fusion, and the arrangement of this research work. Thenintroduces the basic theory of image fusion, including fundamentals, preprocessing, fusionprocessing and quality evaluation.The basic theory of wavelet analysis is introduced in this thesis, and the connections anddifferences between wavelet transform and wavelet frame transform and the characteristics ofmulti-focus images are also discussed. Propose two fusion methods: one is a fusion methodbased on wavelet transform and edge detection, in which the fused coefficients are chosenthrough comparing the edge information of source images according to the waveletcoefficients. Another is a fusion method based on wavelet frame transform and region energy,in which the low-frequency coefficients is divided into focus field and defocus field accordingto the energy intensity of the high-frequency coefficients of each layer, and then partitionedfuse according to the decision table of the focus and defocus fields. The experimental resultsshow the fused image through these two methods can reach nice visual effect and get goodperformance index.The connections and differences between curvelet transform and wavelet transform andthe characteristics of multi-sensor images are discussed. And propose two methods: one is afusion method based on local weighted, in which the low-frequency coefficients are gained byaverage and high-frequency coefficients are gained by local variance. Another is a fusionmethod based on intensity consistency and windows matching, in which the curverletlow-frequency coefficients of the source images are made to be the same mean value andvariance through intensity consistency transform, and then calculate the weighted average ofthese low-frequecy coefficients to be the fused low-frequency coefficients, the fusedhigh-frequency coefficients are calculated in the rule of significance measurement andwindows matching. The experimental results show the fused image with these two methodscan maintain good detail information.