Research Of Generalized Distance Measure And Multi-modal Image Registration

Image registration is the key step in image processing and analyzing. It is requisite for images comparison, data fusion, changes analyzing and objects recognition. The images which need to register each other are always beyond one-modal images. In many cases, it is required to align images taken from different imaging apparatus. In medical image processing, there are many types of images, such as, Tl-weighted magnetic resonance image, T2-weighted magnetic image, proton density weighted magnetic image, functional magnetic image, computerized tomography, single photon emission computed tomography, positron Emission tomography, and B-ultrasonic etc. The integration of these medical images can provide functional and anatomic information. Image registration plays a key role in image clinical application. In remote sensing, there are many types of images, such as, microwave, infrared, radar, multispectral and so on. Registering these images can give more comprehensive information about earth and ecological geological resource. Owing to the importance and perspective of wide application, image registration is catching more and more concern from the world research institutes, and becomes an important and active research area.Multi-modal image registration is a difficult task. From different imaging apparatus, the images of same object in same scene show sharp difference in gray features. In some case, an object displays clear image in one modal, however, it maybe disappears in another modal. On the one hand, there always exists complementary information among Multi-modal images, on the other hand, much common information among images of same body anatomy or same geographic physiognomy. The characteristics of Multi-modal image registration result in feasibility in theory and difficulty in implementation.In recent years, many Multi-modal image registration measures are proposed on the basis of Shannon information theory. These measures have been applied in some research areas. Based information theory measures, in particular, the measures based on mutual information, have been widely accepted in Multi-modal image registrationapplication. However, the methods of mutual information have unconquerable disadvantages, such as heavy cost of count, sensitivity to noise and continuous requirement of probability density function. These shortcomings of mutual information measures restrict their broader practicable application.In the dissertation, a comprehensive classification and summarization is accomplished, and an overall outlook is proposed. On account of the existing problems in Multi-modal image registration, further discussion and investigation have been carried out in this dissertation. On the basis of enough analysis, detailed comparison, and adequate experiment, a novel class of practicable measures is proposed.The innovations in this dissertation can be listed as follows:1. The changing rule of joint gray-value distribution of two images is qualitatively analyzed, according to their relative displacement, such as, independence, incomplete registration, and accurate registration respectively. What is more, an visual conceptual explain is given. The aim and requirement of constructing registration measure is to find a proper function that can measure the cluster of joint probability density function.2. The existing typical statistical multi-modal image registration measures have been comprehensively analyzed and compared in the dissertation. The relation among these methods are anatomized, and through experiments, their performance difference has been experimental is reviewed and remarked. Some improvement has been made for some of these measures.3. A novel idea of generalized distance measures is proposed, and the corresponding formularization is defined. Using the new definition, a novel class of multi-modal image registration measures is constructed afterward. The novel and methods provides more selective extension than information theory, and is validated by experiments. The results of tests show the new generalized di...