Image retrieval method and device based on deep learning

Abstract

The invention relates to an image retrieval method and device based on deep learning. The method comprises the following steps: building a deep neural network; inputting the images of the training setinto a deep neural network for training; loading the trained deep neural network model, inputting all images of the training set into the deep neural network to obtain binary Hash codes with semanticinformation, meanwhile, obtaining the binary Hash codes with visual information by adopting a traditional binary Hash coding method, and establishing a local feature library; inputting an image to beretrieved into the deep neural network, obtaining corresponding binary Hash codes with semantic information, obtaining the binary Hash codes with visual information of the image to be retrieved by adopting a traditional binary Hash coding method, comparing the binary Hash codes with a local feature library, and obtaining a retrieval result by calculating the similarity. Important information of the image can be reserved as much as possible, and rapid and accurate image retrieval of mass image data can be achieved.

Description

technical field [0001] The invention belongs to the application of deep learning in the field of image processing, and in particular relates to a method and device for performing binary hash coding on images by using a deep neural network for retrieval. Background technique [0002] Image retrieval technology aims at the image content that users are interested in, and presents related images to users in a manner of increasing similarity from high to low according to a specific similarity measurement standard. The core problem is how to condense the information of the image, obtain the feature descriptor of the image, and fully express the content information of the image. [0003] The traditional image retrieval technology extracts image features based on basic features such as texture, color, and shape of the image, and uses the corresponding image similarity measurement method to calculate the similarity. However, these basic image features cannot describe the semantic co...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the disadvantage that the existing image retrieval method based on deep learning often adopts a single pooling method in the network design process, the present invention adopts the output structure of each layer in the design process of the network structure Two pooling methods, maximum pooling and average pooling, greatly preserve the semantic information of images to achieve better image retrieval results

Images