Research On Image Fusion Algorithms Using Multiscale Analysis

Image fusion technique means to fuse the imaging information of the same scenewhich acquired by image sensors with different mechanisms and working in differentwavelength into a new image by using certain algorithms, so that the fused image ismore credible, less fuzzy and more comprehensible and is more suitable for humanvision and computer detection, classification, recognition and comprehension. Themultiscale-analysis-based image fusion method which simulate the coarse-to-fineperceptual way of human vision, can obtain a good fusion performance, so that thismethod is always the hot topic in.the research field of image fusion.In this dissertation, the image fusion algorithms based on multiscale analysis arestudied. The main idea and implementation of multiscale analysis are introduced.Based on this method, the main principle of image fusion algorithm is deeply studied.The general frame of fusion algorithm is given, some fusion scheme based on theframe are introduced. The evaluation criteria for the fusion performance are alsostudied in this dissertation, the criteria are summarized according to their property. Anumber of specific image fusion algorithm using multiscale analysis are studied. Fordifferent source images, different fusion condition and different multiscale methods,some specific fusion schemes are proposed. The theory, algorithm structure andimplementation procedures of the proposed algorithms are described in detail.Massive simulation experiments are carried out, and the fusion results are evaluatedand analyzed from subjective and objective aspects. The main contents and contributions of this work are summarized as follows:1. The image fusion system based on multiscale analysis is studied. Certainmultiscale transforms and derivative forms are studied, the general framework forimage fusion is studied, the evaluation criterion of fusion performance is summarized.2. The shift-variance problem of discrete wavelet transform is explored. A novelfusion algorithm based on nonsubsampled discrete wavelet transform is proposed, andexperiments prove that this algorithm is shift-invariant.3. For the imaging characteristics of multifocus images, a window-based fusionrule by using the focus measure function is proposed.4. A novel algorithm based on the uniform discrete curvelet transform is proposedfor the high efficiency, less redundancy, multi-direction and nearly shift-invariance ofthis transform. A hybrid fusion algorithm is proposed using the various advantages ofuniform discrete curvelet transform and the nonsubsampled discrete wavelettransform in representing images.5. The nonsubsampled uniform discrete curvelet transform is proposed. Adenoising and fusion method based on this transform is also proposed, and thismethod can reduce unnecessary transform operation. Aiming at the infrared andvisible images in a same scene, a novel fusion algorithm based on the nonsubsampleduniform discrete curvelet transform and region segmentation is proposed. Thisalgorithm uses the regional contrast ratio and regional standard deviation as thefusion rule of coefficients.