Image retrieval method based on artificial intelligence and related equipment

Abstract

The invention provides an image retrieval method based on artificial intelligence and related equipment. Comprising the following steps: inputting a query image into the quantitative retrieval model to obtain a target category quantitative vector representing a target category to which the query image belongs; obtaining a target complementary feature vector of a triple of the query image under the target category; obtaining undetermined complementary feature vectors corresponding to a plurality of images mapped by the target category quantization vector through hierarchical retrieval; and respectively carrying out similarity measurement on the target complementary feature vectors so as to quickly and accurately screen out a target image meeting the similarity requirement of the query image from a plurality of images under the target category in the image library. Therefore, due to the reduction of the dimensions of the complementary feature vectors and the number and dimensions of the category quantization vectors, the occupation of the storage space and the distance calculation amount are greatly reduced; and distance calculation with a large-scale clustering center is not needed, so that the calculation pressure is greatly reduced, and the invention can be better suitable for retrieval of a large-scale image library.

Description

technical field [0001] This application relates to the field of retrieval technology, and in particular to an artificial intelligence-based image retrieval method and related equipment. Background technique [0002] With the rapid development and increasingly wide application of computer technology, multimedia technology and network technology, the scale of databases / sets is getting larger and larger, such as in information retrieval / recommendation applications, how to retrieve quickly and accurately from large-scale databases The object set required by the application has become a hot research direction in this field. [0003] Taking the image retrieval application as an example, the quantitative retrieval method based on Kmeans is usually used at present, because it clusters the embedding vectors (that is, the embedding vectors, which are also the feature vectors of the corresponding images) of each image in the image library, and the obtained Multiple clustering centers ...

AI Technical Summary

Problems solved by technology

[0004] Moreover, when it is necessary to retrieve several target images with high similarity with the query image from the image library, it is not only necessary to calculate the distance between the embedding vector of the query image and each cluster center in the index, but also to calculate the determined distance The distance between the embedding vector of each image associated with the nearest clustering center and the embedding vector of the query image is used to filter out several target images with small distances, which leads to a large amount of calculation and time-consuming in the entire retrieval process, and additionally occupies the computer. The storage space of the device, which requires extremely high storage capacity and computing power of the computer device, limits the applicable scenarios of this retrieval method

Images