The Study On Multi-sensor Image Fusion In Remote Sensing

As one of the important aspects of image fusion, multi-sensor remote-sensing image fusion (MS-RSIF) investigates how to integrate image information from different aerial remote sensors, and generate new data, for more complete, objective, and essential cognition of the specified target. With the development of information technologies, currently, RSIF becomes indispensable in image processing and understanding, along with many important military and civil applications.In this thesis, MS-RSIF methods are investigated in different levels by means of image processing and modern signal processing techniques. Through fusion of image data from different aerial remote sensors, we can improve the image resolution and analysis results in precision and believability, and further improve the effectiveness in detection and recognition of specific aerial targets.In the following is the main contents and contributions of this thesis:1. Basic concepts and principles of RSIF are introduced, and many traditional algorithms are then systemically analyzed; With detail comparison of their characteristics and performances, a series of useful conclusions are given;2. As for feature-level fusion, a new algorithm based on orientation information measurement and HIS transform is proposed. The experimental results have demonstrated that the proposed method very suitable for fusion of spectrum images in mostly preserving spectrum information and resisting noises;3. As for the multi-resolution image fusion, a wavelet transform based self-adaptive algorithm (DWT_EI) is proposed. Compared with traditional ones, our method has best performance in obtaining information entropy with lower calculation cost, and more correlation with the source image;4. Ten criterion items in four groups based on information quantity, statistic speciality, correlation and gradient respectively, are summarized for fusion evaluation, which are utilized in the experiments to acquire both qualitative and quantitative analysis of the fusion results;5. Not only verified the conclusion that image fusion can improve image information and efficiency of segment, classification and recognition, but also further discussed the recognition and decision problems involved in image fusion. In Addition, reliability and fault-tolerance are also analyzed;6. Except for multi-source remote sensing images, a group of multi-focus images are also applied in our experiments to validate and evaluate fusion algorithms with more intuitional and explicit results.