Research On Image Matching Technology Based On Local Structural Consistency Evaluation

Image matching has always been a hot research topic in the field of image processing and computer vision.Related technologies have important applications in image recognition,image retrieval,target tracking,3D reconstruction and other image understanding and computer vision tasks.The existing image matching research is mainly based on the expansion of image local features,and the initial matching set between the image features to be matched is established by using the characteristics of local features such as representation invariance and identifiability.Considering the large number of mismatches in the initial matching set,the existing research usually introduces the geometric constraints of global or local consistency transformation between real matching pairs,and selects the correct matching in the initial matching set by voting,graph structure optimization or transformation parameter space clustering.However,due to the complexity of image content and the limitation of local consistency quantization and optimization strategy,the performance of existing matching algorithms is still difficult to meet the needs of specific applications.Based on the existing research,this paper proposes a reliability evaluation model based on local structure consistency,and proposes a progressive image matching algorithm around the evaluation method(1)A reliability evaluation method based on local structure consistency is proposed.The observation shows that there are a large number of correct matching pairs in the local neighborhood of the image,and the coordinates of the corresponding feature points on different images are invariant to the topological structure of the image transformation,while the wrong matching does not.Firstly,the confidence of the matching points is estimated to match the neighboring points in the image The difference of the coordinates of the matched feature points shows the structural difference of the local neighborhood of the feature points in different images,and then indicates the confidence degree of the candidate matching pairs.In the specific evaluation process,the main gradient direction of each feature point and the coordinate of feature points are used to establish the linear equation to complete the neighborhood mapping relationship modeling.In the experiment,this method is used to filterthe initial image matching set,and the results verify the effectiveness of the evaluation algorithm.(2)A progressive image matching algorithm based on confidence evaluation is proposed.The algorithm consists of two stages: shrinking phase and expanding stage.In the contraction stage,the confidence evaluation method is used to detect the credibility of each match in the form of iteration,and the low confidence matching is deleted continuously.The goal is to eliminate the interference of low confidence matching evaluation and matching selection process,and finally select the contraction matching set with high confidence.In the expansion stage,the neighborhood matching pairs of the existing selected matching sets are continuously searched in the image space,and the confidence of these matching pairs is evaluated,and the matching pairs with high confidence are continuously collected to expand the existing matching set and improve the recall rate.The matching results on existing standard datasets show that the proposed two-stage progressive image matching method can effectively improve the performance of feature matching.