Research On A Kind Of Multi-scale Image Fusion Method

In recent years, with the development of image sensor, there are more and more ways to obtain images. In order to promote the applications in various fields, the study on the multi-sensor image fusion technology becomes more and more important for scholars. Image fusion is the process of combining images of a scene obtained from multiple sensors to obtain a single composite image. Multiple source images from various sensors can provide either complementary or redundant information. It is possible to utilize the redundant information to improve the signal-to-noise ration and get more reliable results. Likewise, it is possible to fuse the complementary information to obtain a composite image with more detailed and complete content. With the use of multi-sensor (multiple images), the fused image contains a more complete and accurate description of the scene than any of the individual source image. It has been widely applied in many civil fields such as remote sensing, medicine, industry, computer vision and military fields such as scene matching, navigation, guidance, because it has the advantages of improving the reliability, stability and the definition, expanding the space and the time covering scope. Multi-sensor image fusion algorithm and the performance evaluation of image fusion results are two important topics, which are discussed in this dissertation.1. One new multiresolution image fusion algorithm based on probabilistic model is developed. While there are multiple images obtained by different sensors to measure the same object scene, first, multi-sensor vector model of temporal image in multiscale domain is developed, subimage pyramid based on the sensor image is established and the probability model based on the pixel of the subimage at each scale is given. Second, estimate the model parameters using least squares method based on the corresponding pixel of the subimage from each sensor at each scale, then based on the Bayes rules derive estimate of the pixel. Finally, perform inverse transformation to obtain the fusion image. Multiple experiments shows that the fused image obtained by the new algorithm not only possess of lower noise, but also it is better to hold the essential image features of object image than other methods. A quantitative analysis also shows that the new algorithm outperforms the existing algorithm in the case of same condition.2. The performance evaluation of image fusion results is studied. Performance evaluation is a very important issue, it play the guiding role in the image fusion algorithm. In view of information theory, performance indices based on information theory are further analyzed, and then normalized mutual information entropy is proposed successively. The rationality of the proposed performance evaluation index is illustrated by proven and analysed the characters of it. At the same time, the effectiveness of the proposed algorithms is shown through computer simulations.At the end of this dissertation, the further study of image fusion is described and the foreground of image fusion is expected.