The Research On Landmark Extraction And Matching For Remote Sensing Image

Landmark matching is to locate the landmark in remote sensing image using image matching technologies to search over large-scale area. The matching accuracy and efficiency are two key indicators of an algorithm, and most of proposed algorithms are still not match real application. Focusing on accuracy and efficiency of algorithms, this paper proposed several improved algorithms, and the main contribution lies in following three aspects.1. An improved normalized intensity combination matrix correlation algorithm(NIC algorithm) was proposed in this paper. With adopting a similar threshold method, matching happens only when the ratio of intensity summation of image patch and template lies between upper and lower limit, thus the searching area and matching times was reduced. Experiment shows that the proposed method works much more efficient while keeping the accuracy on the same level.2. An edge extraction algorithm improved from Canny operator was proposed. Canny operator is commonly used in edge extraction, but it's hard to tune parameters of Gaussian kernel and thresholds, Also Gaussian filtering is contradictory with edge detection accuracy. To improve the performance, a self-adaptive filtering method is adopted, and the gradient is calculated through a double orthogonal gradient method. Combined with the Otsu method, the thresholds are fixed adaptively too. The experiment results indicate that the improved method is robust to noise and can detect the edge of landmarks more accurate and more complete.3. An improved edge based landmark matching algorithm was proposed with a novel distance transform method. Traditional distance transform method is computational expensive, but with using our edge extending method which get the minimal distance by morphological filtering, the computational complexity was significantly reduced and matching process was speed up. Through the simulation, the improved algorithm can greatly reduce the amount of computation, and improve the speed of matching with a high matching precision.