Study Of Multisensor Image Fusion Methods

In recent years, image fusion has been an important and useful technique for image analysis and computer vision. Image fusion is a technique to combine information from multiple images of the same scene. Multiple source images from various sensors can provide either complementary or redundant information. It is possible to utilize the redundant information to improve the signal-to-noise ratio and get more reliable results. Likewise, it is possible to fuse the complementary information to obtain a composite image with more detailed and complete information content. With the use of multisensor (multiple images), the fused image contains a more complete and accurate description of the scene than any of the individual source images. As a result of this processing, the fused image is more useful for human and machine perception or further image processing tasks such as object recognition and feature extraction. In adverse environmental conditions (such as smoke, fog, rain, low light, motion) or when one imaging sensor may not provide enough information needed for recognition or scene interpretation, we can also get an improved image through image fusion.Multisensor fusion system is characterized by its highly fault toleration. An optimization model and a criterion with minimum cost which fulfilling the system抯 availability constraint are presented for redundant multi-sensor system. In addition, simulated annealing algorithm for the combinatorial optimization problem here is also given. Experimental results show the optimization model, criteria and algorithm are very effectual.In this dissertation, we focus on the problem of multisensor image fusion. The three levels of image fusion (pixel-level fusion, feature-level fusion, and decision-level fusion) are discussed in detail. The simple image fusion method, the multiresolution image fusion techniques based on multiscale pyramid decomposition (MPD), and the image fusion method based on wavelet transform are studied. This dissertation presents several image fusion methods based on MPD, such as image fusion based on Laplacian pyramid (LP), image fusion based on ratio pyramid (RP), image fusion based on contrast pyramid (CP),and image fusion based on gradient pyramid (GP). An image fusion scheme using biorthonormal wavelet transform is also presented. These approaches have been successfully used in image fusion.The fusion rules and fusion operators are very important for image fusion. Different fusion rules are studied in this paper. A novel region-based measure fusion rule is given, and the corresponding fusion operators used for different fusion schemes are presented.Another of the main contributions of this thesis is the objective evaluation of image fusion performance. Quantitatively assessment of the image fusion performance is an important and a complicated issue. Several evaluation criteria are presented in this dissertation. Using these evaluation criteria, we study how the number of decomposition levels, different fusion rules or the size of region for measure influence the image fusion performance. These evaluation criteria are also applied to compare the performance of various image fusion methods.A series of experiments on image fusion are given in this dissertation. We also get some worthy conclusions. The experimental and calculation results show that the fusion schemes and evaluation criteria are effectual.