Image retrieval method and device

Abstract

An embodiment of the present invention provides an image retrieval method and device. The method includes: using a convolutional neural network to extract the features of the target image, and according to the target image feature map output by the last convolutional layer in the convolutional neural network, the same location The total score of pixels in the feature map is used as the class activation map of the target image; each feature map is multiplied by the class activation map and then summed and pooled, and the sum pooled result is combined with the weight of each feature map of the target image Multiply to obtain the intermediate features of each feature map of the target image; according to the target image category probability output by the discriminative layer of the convolutional neural network and the intermediate features of each feature map of the target image, the spatial semantic features of each feature map of the target image are obtained; according to the target The spatial semantic features of each feature map of the image and the spatial semantic features of each feature map of each image to be retrieved are obtained in advance to obtain the retrieval result. The embodiments of the present invention use the spatial semantic information for retrieval, which improves retrieval precision.

Description

technical field [0001] Embodiments of the present invention belong to the technical field of information retrieval, and more specifically, relate to an image retrieval method and device. Background technique [0002] Image retrieval refers to retrieving images similar to a specified image from an image database. Including text-based image retrieval and content-based image retrieval. Text-based image retrieval uses text description to describe the characteristics of images, while content-based image retrieval uses image content semantics, such as color, texture and layout, for analysis and retrieval. [0003] Traditional content-based image retrieval mainly uses handcrafted features, that is, encodes images with mathematical information such as mean and variance to obtain features, and then performs cosine distance or Euclidean distance matching for retrieval. With the achievements of convolutional neural network in image recognition, it is derived to extract the features o...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the problem of inaccurate retrieval results of the above-mentioned existing image retrieval methods or at least partially solve the above-mentioned problems, an embodiment of the present invention provides an image retrieval method and device

Images