Pixel-level Image Fusion Algorithms With Wavelet Transform

The thesis is devoted to studying pixel-level image fusion algorithms with wavelet transform, on this foundation, image-feature-based multiresolution image fusion methods are also explored to a certain extent. The concept, meaning, categorization, principle and development of information/image fusion technology are involved in the thesis, and the general framework for multiresolution image fusion, the performance assessment of image fusion algorithms, and some important existing multiresolution image fusion schemes are also discussed. During the research work, some kinds of image fusiom experiments are performed and the fusion results are evaluated quantitatively and analyzed objectively, so some favorable conclusions are obtained and some creative work are achieved. The organization and detailed work of the thesis as following: Chapter 1 depicts the concept, meaning and categorization of information/ image fusion, and presents some common used sensor in image fusion. Because the research presented in this thesis focuses on pixel-level image fusion, the principle and methods of pixel-level image fusion are explained. Moreover, the chapter introduces the development of information/image fusion technology, then the research situation of image fusion are summarized. And the information of scholars, academic organizations as well as network resources about image fusion are also collected and provided in this chapter. Chapter 2 introduces a general framework for pixel-level multiresolution image fusion presented by Piella[6,92], because this thesis is concerned with multiresolution image fusion methods at pixel level. Moreover, The chapter shows a detailed version of Piella's framework, then several multiresolution image fusion approaches which fit into the framework are analyzed and disscussed. Furthermore, the chapter addresses some existing performance measures in image fusion. Chapter 3 presents a correlated-signal-intensity-ratio-based mutiresolution image fusion algorithm with weighted average fusion rule. First, the multiresolution architectures of the original input images are obtained using the wavelet transform. The composite approximation coefficients are achieved by averaging the approximation of the sources. To the detail coefficients, the algorithm employ a local area signal intensity as the activity measure, and use the defined correlated signal intensity ratio as the match measure, then, the composite detail coefficients are taken to be a weighted average of the detail of the sources. Finally, the fused image is reconstructed using the inverse wavelet transform. The reasoning conclusions and experimental results show that the presented scheme as well as conventional mutiresolution image fusion algorithms with weighted average fusion rule can achieve good fusion performance, furthermore, the experimental results illustrate that the fusion rule of the proposed method is effective. Chapter 4 uses the correlated signal intensity ratio, presents a computationally efficient algorithm of multiresolution image fusion with weighted average fusion rule. Compared with those conventional multiresolution image fusion methods with weighted average fusion rule, the proposed algorithm achieves a considerable improvement in computational complexity. The experimental results show that images achieved by this efficient algorithm are of equivalent or better quality than those obtained from more complex, conventional image fusion methods. Mask and threshold are two important parameters in the presented algorithm, the influences of their various choices on fusion effect are discussed, and the optimising seletion range of the two parameters are expressed qualitatively in the chapter. Chapter 5 presents an edge-based multiresolution image fusion algorithm with weighted average fusion rule. The proposed scheme, which is actually animage fusion method between pixel-and feature-level fusion, uses edge-features of input images to guide the pixel-level fusion process with weighted average fusion rule, and is an attempt to feature-level image fusion approaches. Object edge-features in the fused image achieved by the algorithm are salient, furthermore, the algorithm has the additional advantage of high precision of pixel-level fusion. The experimental results indicate that images achieved by the proposed method have good qualities. Chapter 6 concentrates on drawing some conclusions of the research work in the thesis, and the problems and shortage of current research work are also illustrated, moreover, the chapter discusses the recommendations for further work as well as the prospect of image fusion. Review the thesis, the creative and valuable work as following: 1. The resources and literatures about image fusion technology are collected and categorized carefully in chapter 1. The development of image fusion are also analyzed, the information of scholars, academic organizations and communities as well as academic and network resources about image fusion are also collected and provided in the chapter. 2. When Piella explained the general framework for multiresolution image fusion presented in the literatures [6,92], She mentioned that relative amplitude could be regarded as match measure of detail components, but in whether [6,92] or other literatures, any relative-amplitude-based mutiresolution image fusion algorithm with weighted average fusion rule could not be finded. So inspired partially by the fusion framework proposed by Piella, based on the sense of relative amplitude, the correlated signal intensity ratio is defined in chapter 3, the reasoning conclusions and experimental results show that it can be available of mutiresolution image fusion algorithm with weighted average fusion rule, in which the correlated signal intensity ratio is regarded as match measure. Moreover, good fusion quality can be achieved through the presented algorithm as well as conventional mutiresolution image fusion methods with weightedaverage fusion rule. The correlated-signal-intensity-ratio-based mutiresolution image fusion algorithm with weighted average fusion rule is also based on the image fusion framework proposed by Piella, but compared with those conventional methods, the presented algorithm has lower computational complexity, simpler match measure and avtivity level measure. 3. when chapter 3 shows reasoning for decision map of correlated-signal-intensity-ratio-based mutiresolution image fusion algorithm with weighted average fusion rule, it is also illustrated how to reason the decision map of conventional mutiresolution image fusion algorithm with weighted average fusion rule. The reasoning conclusions provide rational evidence for mutiresolution image fusion algorithms with weighted average fusion rule, and particularly, have special reference to theory of mutiresolution image fusion with weighted average fusion rule. 4. Whether mutiresolution image fusion methods with weighted average fusion rule, or correlated-signal-intensity-ratio-based mutiresolution image fusion algorithm with weighted average fusion rule presented in chapter 3, they are all based on image fusion framework proposed by Piella[6,92].According to the framework, the decision map of mutiresolution image fusion algorithms with weighted average fusion rule must be achieved through combining match measure with activity level measure. It can be finded that, in correlated-signal-intensity-ratio-based mutiresolution image fusion algorithm with weighted average fusion rule, there is strong relation between correlated signal intensity ratio as match measure and signal intensity as activity level measure. So a new idea comes into being that the decision map of mutiresolution image fusion algorithms with weighted average fusion rule is only controled by match measure which also has activity level measure function in conventional methods. The proposed method in chapter 4, which is called computationally efficient algorithm of multiresolution image fusion with weighted average fusion rule, shows that the assumption can be achieved. Simultaneously, based on theidea, a new framework for mutiresolution image fusion with weighted average fusion rule is illustrated in chapter 4. Compared with conventional methods, the proposed algorithm achieves a considerable improvement in computational complexity or real time capabilities. The experimental results show that images achieved by this efficient algorithm are of equivalent or better quality than those obtained from more complex, conventional image fusion methods. 5. In the experiment of computationally efficient algorithm of multiresolution image fusion with weighted average fusion rule, the influences of two parameters, mask and threshold, on fusion performance are discussed in chapter 4. Specially, in the experiment , low pass filter, mean filter and high pass filter are respectively employed as mask. The experimental results show that, to the proposed efficient scheme, the optimising choice of mask is low pass filter. The conclusions can be used for reference by other image fusion methods with parameter of mask. 6. Petrovi?[13,93] presented a scheme which is based on the framework for combining image edge feature with pixel-level maximum fusion rule, partially inspired by the framework proposed by Petrovi?, an edge-based multiresolution image fusion algorithm with weighted average fusion rule is proposed in chapter 5, then it is obtained a framework for combination of image edge information and pixel-level weighted average fusion rule. Strictly speaking, the two methods belong to image fusion approaches between pixel-and feature-level fusion, and are attempts to feature-level image fusion approaches. The experimental results show that images achieved by the proposed algorithm are of better quality than those obtained from the method proposed by Petrovi?. In summery, image fusion is the focus concerned not only in academic study but also in practical application. Image fusion is an integrated technology which encompasses many theories, techniques and tools. Although the research presented in the thesis is limited, the author has done some beneficial attempts in image fusion field. The deep study in image fusion broad to image processing...