Research And Application Of Image Fusion Method Based On Wavelet Transform Technique

Image fusion, which is a major branch of multisensor data fusion, is drawn from information fusion. In recent years, the technology of image fusion has played an increasingly important role in modern aerospace, automation, remote control and medicine, particularly in the field of military guidance.With the help of complementary data information of multisensor, a new image is reconstructed by image fusion, which provides richer visual information than the original images. The research on image fusion has a lot of aspects, such as fusion algorithms, fusion rules, fusion quality evaluation and the types of images. According to the work space domain, image fusion can be divided into space domain, spectral domain, frequency domain and the scale domain. According to the different abstract extent of information, from low to high, it can also be divided into three levels: pixel-level, feature-level and decision-making level. Pixel-level fusion, as a fundamental fusion, is the basis of the other two, and therefore becomes one of the hottest spots. In this thesis, our study is aimed at pixel-level image fusion.Under different classifications, image fusion algorithms can consist of space domain and transform domain, or pixel-level, feature-level and decision-making level. Today's hot issues of image fusion focus on multisensor, multiresolution and multifocus fusion. The decomposition and reconstruction of multiresolution image fusion can be classified into two categories: the decomposition and reconstruction of pyramid-type and wavelet-based transform (ie Mallat fast algorithm). At present, from the characteristics of the human eyes and visual observation, multiresolution architecture, in the area of multiple grayscale image fusion, is widely accepted as a suitable tool to process the corresponding pixel-level fusion. With further study on wavelet transform, wavelet multiresolution analysis, as a perfect mathematical tool, has been found more and more applications in image processing. This thesis concentrates on the multifocus image fusion, using multiresolution image structure from wavelet transform.Pesudo-Gibbs phenomenon can be effectively avoided by shift-invariant wavelet transform, and with such ideas, Cycle Spinning algorithm, used in de-noise, was produced. But this method is seldom mentioned and used in image fusion literature at present. In this thesis, the method is applied to image fusion, and makes some foundation and expansion researches on image fusion based on wavelet transform using Cycle Spinning.The main contributions of this thesis are listed as follows:1) It investigates a certain number of domestic and foreign periodicals and literatures, overviews the basic concept of image fusion, summarizes the recent progress in this field, discusses the existing problems and outlines the development tendency in the future.2) It makes a detailed classification of image fusion algorithms, and explains space domain, transform domain, pixel-level, feature-level and decision-making level image fusion algorithms specifically, as well as the range of application and respective characteristics.3) It analyses the hot issues in image fusion, and illustrates multisensor, multiresolution and multifocus image fusion respectively.4) Having reviewed the development of the Fourier transform and wavelet transform, it introduces two category methods from multiresolution analysis: the decomposition and reconstruction of the pyramid-type and wavelet-based transform. Then this paper analyses various factors affecting fusion effect: the types of image, base functions of wavelet transform, fusion filters, decomposition levels, as well as fusion rules.5) It discusses evaluation measures of fusion quality, makes a separate list of the combination with subjective and objective estimation method, and uses Graphics User Interface (GUI) and Wavelet Toolbox in MATLAB, to achieve the simulation result visualization of fusion assessment. This work has been published in a paper "Integrated Comparative Research on Evaluation Methods for Wavelet Transform Based Image Fusion."6) It describes the shift-invariant wavelet transform with suppression of Pesudo-Gibbs phenomenon, which has been analysed, employs Cycle Spinning algorithm with the thought of "shift and average" in image de-noising, proposes an image fusion approach based on wavelet transform using Cycle Spinning (called CSDWT, CS for short), makes the realization of simulation process, and gains better fusion of the subjective and objective effects. This work is published in a paper "Studying Image Fusion Based on Wavelet Transform with Cycle Spinning."7) It continues to study Cycle Spinning algorithm, discovers its two existing problems: shift direction and shift quantity of fusion image, and thus raises different shift strategies to improve the CSDWT. The simulation results show that, the fusion effect is not weaker, and such improvements reduce the amount of computation. This research has been published in a paper "Further Study on Image Fusion Based on Wavelet Transform Using Cycle Spinning."...