Studies On Multi-Spectral Image Registration And Fusion

Following their rapid growth, multi-spectral imaging and multi-sensor sensing techniques, have found wide application in many domains such as: remote sensing, anti-terror, military object trace and recognition, flying navigation in night, etc. The applications of these techniques normally result in a great amount of data. The massive data give people a challenge during its processing, analysis, compression and transmission. Multi-spectral image fusion is rising as a new image processing and analysis method which results from this background. Multi-spectral image fusion is combination of the information acquired from multi-spectral sensor in same scene such that clearer description being acquired taking advantage of those correlation in time and space and complementary each other. And it can also reduce the quantity of the data to a extent. In this paper we study multi-spectral image fusion algorithms and image fusion evaluation methods:· HVS theory is introduced into multi-spectral image fusion. And we propose two image fusion method based on HVS theory. One is in space-domain; the other is in wavelet-domain. These two new methods consider sufficiently the property of human being vision. They can improve effectively the contrast of fusion image and bring the targets in image into prominence. So their fusion results are favorable to recognize by eyes. Many experiments show that our methods have some advantage on vision compared with other image fusion methods.· We survey four used commonly fusion evaluation criteria. By using these evaluation criteria we make a quantitive comparison among five image fusion methods include the HVS theory based methods proposed in Chapter 2 through many experiments. In these experiments, we analyze the image fusion methods, and at the same time the evaluation criteria are also evaluated. Our experiments show the quality of fusion results cannot be well evaluated in some instances by these evaluation criteria.· We propose linear optimal and optimal fusion methods under correlation coefficient criteria. And we prove that linear optimal fusion method is equivalent to optimal fusion. The computational complexity of the optimal fusion is greatly reduced derivation from this conclusion. Furthermore, we make some extension of the optimal fusion method so that it can apply to various situations.Before performing image fusion, the images to be merged must be registered. But in practice it usually is not meted. So the images need to be registered. In fact, image...