Research Of Automatic Registration Method For AVHRR Remote Sensing Image

"Remote Sensing Image Registration" is also called " Remote Sensing Image Geometric Correction ". By spatial resolution and observation scale, remote-sensing images can be classified into high-resolution, medium resolution and low-resolution images. Owing to its strong macroscopic characteristics, low-resolution images have been widely applied to weather detection, vegetation cover, land use rate monitoring and other sectors within a continental scale. Among all low-resolution remote-sensing images for vast-range observation, AVHRR image is commonly used. However, for reasons including sensor angle, remote sensing platform attitude and topographic relief, it is unavoidable to cause imaging distortion when AVHRR is used for imaging. Remote-sensing image registration has therefore become an essential step for the subsequent fusion, classification and other integrated utilization.Generally at present, dominating point pair is extracted at automatic mode in registration. So far, dominating point pair extracting has been done mostly based on gray matching and feature matching. Registration based on features is the most widely used. The key to this registration technique is the determination of homonymous dominating point pairs. This paper focuses on feature-based registration and proposes a solution to the problems in automatic registration/ geometric correction of AVHRR image.This paper mainly focuses on two problems in AVHRR geometric correction: 1. there are strong cloud cluster disturbance in AVHRR imaging; 2. there is usually geometry deformation in regions with dramatic topographic relief.To address these two problems, this paper proposes the method of automatic registration for AVHRR images. First of all, on the basis of coarse localization of geographical coordinates, OSTU algorithm is used to tag the clouds and SIFT algorithm is used to extract feature points automatically; then DEM data is used to eliminate potential false feature points in regions with dramatic elevation fluctuations. Based on the DEM data, multiple rectangular partitions with desired internal elevations are generated and false homonymous dominating point pairs are eliminated inside the partitions with RANSAC algorithm. At last, with the AVHRR L1 B images got by the National Satellite Meteorological Center as the experimental data, the dominating point pairs are processed with geometric correction. Experiments show that large low-resolution remote-sensing images can be processed effectively with the method; meanwhile, it significantly increases the precision and efficiency of geometric correction of AVHRR image.