Research And Application Of Corner Detection Algorithm Based On Multi-layer Wavelet Decomposition

The Remote Sensing(RS)data pre-processing is an indispensable step for product production,its processing accuracy and speed will have the cumulative effect to follow steps,so we need to ensure the pre-processing high-speed and effective.Image registration as a fundamental procedure of the pre-processing,the geometric feature extraction of image has higher requirements on the precision of geometric feature points.Although there have been some achievements in research of RS image geometric feature extraction all over the world,in case of multi-source RS data,the accuracy of the extraction geometric feature points has not been solved satisfactorily because the image is restricted by noise and lack of features.Meanwhile,after image registration,the standard tile data fast view can be generated based on the Five-Layer Fifteen-Level(FLFL)data organization structure.Imbalance color problem of quick-view caused by tile data processing methods and the number of tiles lead to production efficiency is low also needs to be resolved.In this paper,a corner detection algorithm based on wavelet decomposition is proposed,which improves the accuracy of the control points by extracting high frequency components.At the same time,we use the technology of Web Service and multi-thread to improve the production of RS data.The main contents of this paper are as follows:1.There are many pseudo-corner points in the corners generate by Harris corner detection algorithm,this paper proposed a corner detection algorithm based on multi-layer wavelet decomposition in order to solve this problem.2.In the initial prototype system,the reference image library required for extract the control point is too large,which leads to the problems that the system portability is poor and the image library is not easy to manage.This paper proposed the strategy of on-demand dynamic acquisition for the reference image in order to solve these problems.3.Aiming at the problem of chromatic aberration between different quick-view of different tile data,a method for generating quick-view using image histogram information is proposed.This method can balance the color of quick-view and ensure the ground target clarity,also improved the efficiency ofquick-view production.Experiment data shows that,the modified corner detection algorithm proposed in this paper reduced the average error of the control point by 21.8%;the strategy of on-demand dynamic acquisition can save up to 95% storage capacity of reference image library;introduction of multi-thread processing mechanism made the RS data pre-processing efficiency increased by 18.6%.These above research algorithms and contents have been applied to the RS data pre-processing prototype system in the national major project for high resolution of “High Resolution Earth Observation System".