Image target searching method and device

Abstract

The invention discloses an image target searching method and device. The searching method includes: targeting and locating the to-be-searched image; classifying and extracting local features of the target, to generate the local feature codes for the local features; searching for similar images in the database that belong to the same class with the aforementioned target; comparing the saved local feature codes of the similar images with the codes of the target local features, then outputting similar images with similarity greater than the threshold. Through targeting and locating the to-be-searched image, classifying the target into at least two classes, and generating the local feature codes for the targets. When searching, search according to the search target classifications and order according to the similarities. Through target classifications, the searched target is matched using a model under the same target classification, thus increasing the search accuracy and efficiency.

Description

technical field [0001] The invention relates to the field of graphics and images, in particular to a method and device for retrieving objects in images. Background technique [0002] When retrieving pictures in the prior art, technically speaking, the grayscale image of the target is generally processed, the image local description (SIFT, SURF) is extracted, and the bag of visual words (BOW), Hamming embedding, local sensitive hashing, etc. are used. method to calculate the distance or similarity of features between target images. This method ignores the equally important global features such as color, shape, texture, and target type, resulting in the retrieval result of color, shape, texture, target type and query goals vary widely. [0003] From the perspective of retrieval ideas, the types of targets that are not distinguished in the monitoring scene are compared one by one, the processing is not targeted, and resources are wasted. Contents of the invention [0004] T...

AI Technical Summary

Problems solved by technology

[0002] When retrieving pictures in the prior art, technically speaking, the grayscale image of the target is generally processed, the image local description (SIFT, SURF) is extracted, and the bag of visual words (BOW), Hamming embedding, local sensitive hashing, etc. are used. method to calculate the distance or similarity of features between target images. This method ignores the equally important global features such as color, shape, texture, and target type, resulting in the retrieval result of color, shape, texture, target type and query goals vary widely

[0003] From the perspective of retrieval ideas, the types of targets that are not distinguished in the monitoring scene are compared one by one, the processing is not targeted, and resources are wasted

Images