Research On Image Fusion Technologies Based On HMT Model In The Contourlet Domain

Image fusion technology combines the important information extracted from several source images, which may be obtained by different sensors in the same scene or by the same sensor at different time intervals, into a fused image. The fused image can be used for more efficient analysis and processing of the scene, since it contains more comprehensive and accurate information. Image fusion is a multi-and inter-discipline technology, which covers theories of several fields such as sensor technology, image and signal processing, artificial intelligent, computer technology, statistical and estimation. The concept of image fusion was proposed in late1970s, and the technologies are now wildly applied in fields like battlefield monitoring, digital imaging, medical diagnosis and remote sensing, to name a few.The main research content and results are listed below:(1) The theories and technologies of multi-source image fusion are summarized and categorized. Basic algorithms for fusion and transformation rules in pixel level are described systematically. Assessment methods of the fused results and the main performance indexes are briefly introduced.(2) The theories of several classical image decomposition algorithms, such as Laplace Pyramid Decomposition, Contrast Pyramid Decomposition, Wavelet Transformation and Contourlet Transformation are analyzed. These algorithms have been tested on the fusion of Computer Tomography (CT) and Magnetic Resonance Imaging (MRI) images of human brains. The test results show that, Pyramid Decomposition based algorithms can discriminately highlight the important features and details of the images in multi-dimensional and frequency domain, while the direction information is partially lost which results in blurs; except for the aforementioned advantages, Wavelet Transformation possesses certain dimension selection ability, orthogonality and the time-frequency resolution can be variable, however, the transformation can only capture information at three directions, the special detail information cannot be efficiently enhanced; Contourlet Transformation is multi-dimensional and multi-directional, and its anisotropy ability leads to better extraction of the edges, outlines and direction information than the aforementioned two methods.(3) Provide a multi-focus image fusion method based on wavelet domain HMT model. On the basis of the statistical characteristics of the Wavelet transformed images, the fusion methods for multi-focus images with Wavelet domain Hidden Markov Tree (HMT) model are discussed. Markov chain, Hidden Markov Model, Wavelet domain HMT modeling method are studied in depth. Fusion algorithm based on Wavelet domain HMT model (w-HMT) is applied to multi-focus images, and its effectiveness is verified.(4) Provide an image fusion method based on contourlet domain HMT model. A Contourlet Domain HMT is established after analyzing the probability distribution and parameter correlation of Contourlet Transformation, according to its multi-directional and dimensional features. The fusion methods and rules based on the model are proposed and verified by the tests on multi-focus images. The standard deviation, average gradient and information entropy of the C-HMT fused images outperform that of w-HMT, and the calculation is substantially accelerated at the same time, which indicates the effectiveness of the proposed approach.