Image retrieval method and system based on global and local feature rearrangement

Abstract

The invention relates to an image retrieval method and system based on global and local feature rearrangement. The method includes: extracting the global features and local features of the image to be queried and the images in the image database; calculating the similarity between the image to be queried and the images in the image database according to the global features, and sorting the images in the image database according to the similarity , take the topN image of the sorting result as the query result of the global feature; use the local feature to calculate the matching degree between the image to be queried and the topN image, rearrange the topN image according to the number of matching points, and obtain an accurate image retrieval and sorting result. The present invention first uses the similarity of the global features to perform image retrieval and sorting, and then rearranges the TopN results based on the feature matching points of the local features, further improving the retrieval accuracy, and is especially suitable for the key points of the suspect vehicle in the public security arrest or traffic accident Troubleshoot.

Description

technical field [0001] The invention belongs to the technical fields of information technology and image retrieval, in particular to an image retrieval method and system based on global and local feature rearrangement, and is especially suitable for retrieval of images such as vehicle images. Background technique [0002] In image retrieval methods such as vehicle image retrieval methods, they are roughly divided into two categories: one is based on global features, uses residual network to map vehicle images into feature vectors, and uses the cosine similarity of feature vectors to complete image retrieval sorting. The method only expresses the content of the image, but cannot express the spatial position information of vehicle components, resulting in low average retrieval accuracy. The other is based on traditional local features, mapping vehicle images as key point descriptors, and calculating the number of matching points based on feature matching to complete image retr...

AI Technical Summary

Problems solved by technology

[0002] In image retrieval methods such as vehicle image retrieval methods, they can be roughly divided into two categories: one is based on global features, using the residual network to map vehicle images into feature vectors, and using the cosine similarity of feature vectors to complete image retrieval sorting. The method only expresses the content of the image, but cannot express the spatial position information of the vehicle parts, resulting in low average retrieval accuracy

The other is based on traditional local features, mapping vehicle images to key point descriptors, and calculating the number of matching points based on feature matching to complete image retrieval and ranking. The image retrieval task failed to complete a good expression, resulting in a relatively low average retrieval accuracy

Images