Combined query image retrieval method based on multi-order adversarial feature learning

Abstract

The invention discloses a combined query image retrieval method based on multi-order adversarial feature learning, and the method comprises the steps: firstly obtaining image features through a pre-trained feature extraction module, obtaining text features through an LSTM network, and then fusing the features of two modes through the guidance of self-attention; in addition, generating high-order features from the low-order features in a bilinear fusion mode; learning a similarity relationship between the features by utilizing triple loss, further promoting fusion between the features by utilizing a discriminator and retrieval network confrontation, and finally, training a model in an end-to-end manner by combining the two features, thereby realizing efficient combined query image retrieval. According to the method, the deep learning technology is utilized, the game thought is used for reference, and the performance and efficiency of combined query image retrieval are improved to a great extent.

Description

Technical field[0001]The present invention relates to the field of machine learning combination query image retrieval technology, and more particularly to a combination of multi-order counterfeit characteristics.Background technique[0002]With the rapid development of information technology, mobile networking equipment is applied, and people can easily touch the massive and diverse image resources on the network. In the face of such a huge data, the efficient and accurate image retrieval method and system have become an essential requirement when it is in compliance with the picture. The total number of pictures, which brings large-scale growth of similar images, resulting in the accuracy of the retrieval, and existing image retrieval technology faces huge pressure and challenges. The mainstream text retrieval image and picture retrieval image have respective limitations. If the expression ability of plain text is limited, there is a loss of information when the idea changes into lan...

AI Technical Summary

Problems solved by technology

The current methods also have some shortcomings, they do not make full use of multi-scale features, but the features of different scales often contain information unique to each level

Their fusion methods for the features of images and texts are also relatively simple, and the retrieval efficiency is low.

Images