Research On Multi-source Image Matching Method Considering Color Invariance

Image matching is the most critical step in 3D reconstruction of photogrammetry.The quality of image matching will directly affect the quality of the final 3D reconstruction.With the continuous development of information technology,the image matching technology,which has always been a research hotspot,has also developed rapidly.The development of photogrammetry technology makes the application of photogrammetry 3D reconstruction technology more and more extensive,and the field of cultural heritage protection is one of them.As UAV technology continues to mature,3D reconstruction based on multi-source image matching has become an indispensable part of the cultural heritage and archaeological field.The purpose of multi-source image matching is to find feature points of the same name in the close-range and oblique images,and match two movies through these same feature points.However,in the multi-source images obtained from the site,there are special textures such as repeated masonry textures and poor sand textures.There are difficulties in matching such images using gray-scale image matching methods.For example,when matching poor texture images,there is a problem that no effective matching points can be obtained due to the interference of noise;while matching repeated texture images,due to the ambiguity of feature description,multiple matches will occur Case.In order to solve the above problems,to achieve the ultimate goal of matching the multi-source images of the site.In this paper,the existing image matching algorithm is studied,the most suitable algorithm is selected for improvement,and the method of multi-source image matching based on the improved algorithm is studied.The main research contents are as follows:(1)Comparison of matching effects of common image matching methods under the affine frame.Due to the large affine transformation between multi-source images,many commonly used matching algorithms do not have the ability to adapt to affine transformation.Therefore,it is necessary to build an affine frame with a certain generality,through which the adaptability of the affine transformation of common matching algorithms is improved.Based on the matching experiment results,the improved algorithm is compared and analyzed with the evaluation criteria such as the number of matching points,the distribution of matching points,and the matching time,and the algorithm with better matching effect and room for improvement is selected to carry out the follow-up research on the improvement of the feature descriptor..(2)Enhance the uniqueness of the descriptor based on the image color information.The special texture image features described based on the gray value descriptor are poorly distinguishable,resulting in an error in image matching.Therefore,the richer color information in the image is used instead of gray to describe the features.The color information is introduced into the structure of the MROGH algorithm feature descriptor through the color invariant model to enhance the uniqueness of the MROGH algorithm feature description.(3)Research on multi-source image matching method based on improved algorithm.Study how to improve the MROGH feature descriptor as the core to match the multi-source imagery of the site.This article studies and explains from the aspects of image color preprocessing,determination of feature areas,construction of color gradient histograms based on improved descriptors,and selection of feature fine matching methods.(4)Contrast experimental research on algorithm before and after improvement.By comparing the method in this paper with the pre-improved algorithm in terms of the number of matching points,matching accuracy,matching accuracy and matching time,the experiment proves that the multi-site image matching based on this method can obtain the number of matching points that meet the requirements of image matching,And the matching accuracy has reached the sub-pixel level.