Research On Optimal Image-Fusion Algorithm

This thesis mainly researches algorithms of optimal image fusion from three aspects and makes further research on their evaluation.Aiming at the problems of those existing image fusion methods base on wavelet transform, an optimal algorithm of image fusion based on wavelet transform is proposed. The method starts from the source images'characteristics and analyses the purposes of image fusion, then pertinently selects the proper wavelet base and its fusion rule by studying the fusion effect of different wavelet bases. Thus, the fusion effect is improved.Aiming at the characteristics of infrared and visible light image fusion, this paper brings forward an optimal image fusion method based on object extraction. After extracting the object from infrared image; the method fuses the extracted result with visible image to make full use of the source images.Aiming at the problem of wavelet transform in the image's edge representation, an approach based on curvelet transform is also studied in this paper which makes a good fusion result with abundant information and better edge representation. In addition, aiming at the problems of single-factor evaluation on the effect of image fusion, which only considers some feature of the fusion image and lacks the global characteristic, a comprehensive evaluation method based on artificial neural networks is proposed. Entropy, spatial frequency, cross entropy and mean-square error of the image is selected to form a set of single-factor evaluation index. After getting the evaluation value of single-factor index through fuzzy integral, the method takes them as inputs of artificial neural network to train their weights. Thus the comprehensive evaluation index is achieved.The experiments in this thesis prove that the algorithms and methods mentioned in this paper can effectively improve the effect of image fusion.