Image retrieval method and device and image representation network training method and device

An image retrieval and characterization technology, applied in the computer field, can solve problems such as the influence of user experience, the increase in calculation and time consumption of image retrieval, and the difficulty of image retrieval technology to meet the real-time requirements, so as to reduce the calculation time and the time consumed by retrieval and computing resources, the effect of improving retrieval efficiency

Active Publication Date: 2021-08-13
ALIPAY (HANGZHOU) INFORMATION TECH CO LTD
View PDF7 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, as the number of images in the retrieval database continues to increase, the amount of computation and time-consuming image retrieval increases, and existing image retrieval technologies are difficult to meet real-time requirements, resulting in affected user experience

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Image retrieval method and device and image representation network training method and device
  • Image retrieval method and device and image representation network training method and device
  • Image retrieval method and device and image representation network training method and device

Examples

Experimental program
Comparison scheme
Effect test

Embodiment approach

[0056] For the image representation network mentioned above, in one implementation manner, it can be trained by means of metric learning. In one embodiment, pairs of image samples of the same type and pairs of image samples of non-types can be constructed as training data, and the image representation network is trained by calculating triplet loss.

[0057] In addition, the inventor considers that, in the image retrieval method disclosed in the above embodiments, the image retrieval is realized by using the binarized feature vector. However, the vector distance obtained by comparing the same position elements with the binarized feature vector is not continuous everywhere, so it cannot help optimize the image representation network during the training process. Further, the inventor found through mathematical demonstration that in a discrete space composed of values ​​-1 and 1, there is the following linear relationship between the Hamming distance and the cosine similarity:

...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The embodiment of the invention provides an image retrieval method. The retrieval method comprises the following steps: acquiring a first image to be retrieved, and inputting the first image into a trained image representation network to obtain a first representation vector; carrying out binarization processing on the first representation vector to obtain a binary representation vector of the first image; on the other hand, obtaining a plurality of binary representation vectors of the plurality of candidate images; and further, respectively calculating vector distances between the plurality of binary representation vectors and the binary representation vector of the first image, and recalling an image similar to the first image from the plurality of candidate images based on the vector distances. The embodiment of the invention further provides a training method of the image representation network. The training method can ensure that the retrieval performance of the image features before and after binary processing is kept consistent.

Description

technical field [0001] One or more embodiments of this specification relate to the field of computer technology, and in particular to a method and device for image retrieval, and a method and device for training an image representation network. Background technique [0002] Image retrieval is involved in many fields. For example, search engines provide image search functions, support users to input images for search, and return the retrieved identical or similar images to users. However, as the number of images in the retrieval database continues to increase, the amount of computation and time-consuming image retrieval increases. Existing image retrieval technologies are difficult to meet real-time requirements, resulting in affected user experience. [0003] Therefore, there is an urgent need for a solution that can effectively reduce the time delay of image retrieval, thereby improving user experience. Contents of the invention [0004] The embodiment of this specificat...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06F16/583G06F16/55
CPCG06F16/583G06F16/55
Inventor 唐董琦
Owner ALIPAY (HANGZHOU) INFORMATION TECH CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap