Studies On Image Fusion Based On Wavelet Transforms

Image fusion is one of the important parts of multi-sensor information fusion. It aims at getting more exact and more comprehensive image description of the same scene or object by distilling and integrating the information of multi-source images. Multi-sensor image fusion differs from the general image enhancement. It is a new technology in areas of computer vision and image understanding.This dissertation is mainly focused on the ways of image fusion based on pixel level. At first, we introduce the background, the development conditions of the image fusion and its search conditions; review the concept, lever of image fusion, three means in common uses and the theory foundation of the wavelet transforms.Then, we introduce a kind of image fusion method based on region — segmentation which is the research emphasis of this dissertation. We firstly carry out multi-resolution segmentation based on pyramid linking by considering all source images. The segmentation is corresponding to the sub-images reduced by the source image's wavelet transforms. Then the segmentations are fused according to some regulation. Lastly the result image is obtained by means of inverse wavelet transforms of the fused sub-images. We use shaped rational cubic spline functions to dynamically calculate fusion operator. In this way different fusion operators can be chosen according to different purposes for the perfect effect, which is proved to be valid by a mass of experiments.At last, we introduce the character of multi-focus images based on definition. The definition of image is introduced in the dissertation via eight-neighborhood Laplacian and is used to fuse the multi-focus images. Experiments show that good effect can be achieved with this fusion method.