Study On The Key Technology Of Image Fusion Based On Multiresolution Analysis

As one of the modern high technologies, image fusion is a combination of sensor,image processing, signal processing, computer and artificial intelligence. Through thistechnique, the complementary and redundant information of multiple images can bemerged to make up the shortage of single equipment. Simultaneously an image whichcontains more accurate, reliable, comprehensive information can be obtained for theconvenience of human visual perception or for further tasks, such as processing,analysis, decision and so on. Though image fusion has widely used in military, medicalimage processing, remote sensing, multi-sensor network, and other areas, it is still ofgreat significance in studying the key technology of image fusion systems, because thedomestic and foreign related research is not adequate in theory and applications, and thekernel technologies are still blockaded by the developed countries.The thesis researches the key technology of image fusion based on multi resolutionanalysis, puts forward some new ideas and methods of image fusion and qualityassessment. The main work is as follows:In the foreword, the background of the problem is introduced, and the developmentof fusion technique in our country and abroad was investigated. The second chapterinterprets the frame work and categories of image fusion, and gives the frequently usedimage quality evaluation indicators, and summaries the key influencing factors of imagefusion based on multi-resolution analysis.Then, the thesis makes a comprehensive comparison and analysis of image fusionbased on multi-resolution analysis, and obtains some useful conclusions in the thirdchapter. First of all, it introduces the principle of multi-resolution analysis and imagefusion based on it. Then it elaborates the fundamental and features of pyramid transform,wavelet transform and shift invariant wavelet transform, especially the merits anddrawbacks of each method are also illustrated in detail. Furthermore, to test anddiscover the influence of different multi-resolution analysis methods, differentparameters and different levels of DWT, the thesis designs four simulations, andevaluates the results with various objective and subjective image quality indicators.Through this way, some rules in multi-resolution analysis image fusion are revealed,and some conclusions about how to choose the parameters and what aspects shouldgiven more attention to are drawn. These works are very useful in the study of the samefield. Next, the thesis puts forward a novel image fusion method based on theNon-subsampled Contourlet Transform (NSCT) and multi-scale statistic for multi-focusimage fusion in the fourth chapter. The objective of multi-focus image fusion is tocombine the clear part of different source images with the same scene to a single image,and how to get a better focus measurement is the key to the question. The papermanaged to analysis the inputs image in NSCT domain with multi-scale statisticalfeatures and gives a novel fusion strategy to select the coefficients. The multi-scaleaverage gradient is applied on the low frequency coefficients. And all of the highfrequency coefficients of the same position were selected from the same source imagedecided by the average of the multi-scale energy value of these subbands. Thecomparative result with the other methods based on Wavelet Transform (WT),Contourlet Transform (CT) and NSCT shows that the proposed method exacts the mostinformation from the source images and achieves better performance both in thesubjective visual perception and quantitative metrics.Then, considering of remote sensing applications, the paper proposed an imagefusion method based on region feature and NSCT to resolve the problem of Spectraldistortion. By discussing the characteristic of PAN and MS images, a new fusion rulewhich divided the input image into three kinds of regions according to the salience oftheir NSCT coefficient was presented. And for different kind of regions different fusionalgorithm were adopted. The paper also improved the assessment method the remotesensing image fusion, and two new image quality indicators named the ratio of spectralmaintained and the ratio of average gradient were put forward which can give betterunderstanding of the fusion image quality compared with the inputs. Moreover, thepaper go deep into the determining parameters of the region salience, and give therelation between the parameter value and the final fusion image qualities, through withsome useful conclusion were drawn.After, the sixth chapter presents a novel objective image quality assessmentmethod for image fusion based on mutual information and structural similarity. Asimplified formula was deduced to compute the information amount transformed fromthe source images to the final fused image. With this formula we managed to resolve theoverlapping information problem immediately and predigested the calculationextremely. Moreover we adopted the mutual information and the structural similaritymetric to image fusion schemes, and put forward a novel performance evaluationmethod without the interference of the reference image. Experimental results show that the proposed metric is well correlated with human subjective evaluations and thus canbe used to compare the results of different fusion methods.Finally, the thesis over reviews the main work of the paper, and gives the forest ofthe development of image fusion.