Multi - Sensor Image Splicing, Fusion And System Implementation

With the rapid development of sensor technology, more and more be used to complete different function of sensors are applied to each field, the traditional single sensor model evolved into multi-sensor. Because the multi-source data of different sensors are complementary and redundancy, therefore multi-sensor model can provide a more comprehensive, rich and reliable target observation information. Mosaicing and multi-sensor data fusion method, the author of this paper was studied, the main research work and research includes the following several aspects:(1) In this paper, an image mosaicing and fusion system is designed and implemented based on visible, infrared and hyperspectral images. The system contains sub-images mosaicing from the same sensor in low coincidence rate, image registration and fusion between different sensor with high precision, as well as other parts of the auxiliary functions.(2) Because the sub-image size is 4 x 4 and the overlap rate between two adjacent images must be less than 2%, this paper proposes a fast equidistant matching of randomly selected points image mosaicing algorithm. The algorithm moves compared area with same step between two adjacent images. Taking randomly selected points in this area and calculating the color difference between the corresponding points, when the accumulated value exceeds a certain threshold value, the calculation is stopped. In the process of comparing calculation, the most frequently compared regional center position is the best matching position. Meanwhile, a template based mosacing algorithm is also realized in our system as substitution.(3) Using image registration algorithm based on SIFT feature points for registration between infrared and visible images; The registration between infrared and hyperspectral data, a registration algorithm based on Harris corner and Hausdorff distance is proposed. The system can properly fulfill both two image registration task verified by actual image.(4) Related algorithm is improved based on the analysis of visible, infrared and hyperspectral image characteristics, then fuse multi-sensor images. For variational PDE method, we use variation model based on gradient descent, and take fusion experiments; For multi-resolution analysis, we use wavelet and Contourlet transform based on regional correlations. The experiments show that the fusion algorithm can meet the system requirements.