A Study On The Strategies Of Close-range Trees Image Matching Based On Feature

With the computer stereo vision technology continues to progress and development of close-range photogrammetry gradually, the close-range images as the basis of stereo vision can be reproduced possibly. In the premise of not use high-precision measuring instruments, getting the trees images only by non-measurement digital photo, and then restoring its true three-dimensional shape has the low cost, application flexibility and other characteristics, so it has become the subject that more and more researchers compete for.This study is just based on these considerations, and attempts to use the low-cost, easy operation non-measurement digital camera as a tool, get the close-range form of trees through processing the obtained images, and designs reasonable strategies for image matching. So that it can provide the most direct data base for the visualization of trees and the build of stereo vision models.Based on the analysis that, compared with the related methods of regional gray-based images, feature-based images matching method takes into account the differences or distorted exist between the relevant images, and also can get a better matching results in a region that has brightness uniformity and poor information. Therefore, constructs the feature-based image matching strategies for the trees. The basic process is:image acquisition; image processing; image segmentation of trees; feature extraction; image matching.This article studied the image processing methods for the purpose of image segmentation firstly and feature extraction, mainly from both the frequency domain and space domain directions enhanced on the original image, to make it more suitable for computer recognition and more prominent target. Then the algorithm for image segmentation of the tree and feature extraction is one of the focuses of this study, a good image segmentation method determined the effect of feature extraction, while the feature definition and feature extraction method provide the basic objects for the realization of image matching.This paper through the analysis of the threshold, region growing, cluster classification and other basic segmentation algorithms, combined the core strengths of various algorithms and measurement methods, designed and implemented two trees image segmentation methods that based on clustering thought, and using Moravec operator and the SUSAN operator to get the extraction of feature points from the original image and after the segmentation of images respectively. By comparing the extracted results, the feasibility of the two segmentation algorithms is confirmed. The most important part of this article is for the close-range images of trees, designed and implemented two feature matching process-oriented strategies and analysis of the results of matching with a reasonable accuracy evaluation method.Experimental results show that the design of the two images based on feature points matching strategy of trees can achieve better results and higher matching accuracy; the scale-invariant features could be used as matching primitives, which is more reasonable in theory, and the matching results in the precision and accuracy are better than matching method that simply to match the characteristics of gray as matching primitives.