Image retrieval-oriented deep reinforcement redundancy elimination Hash algorithm

An image retrieval and depth technology, applied in still image data retrieval, still image data query, reasoning methods, etc., can solve the problems of hash code including redundancy, low sampling efficiency of hash algorithm, inability to maintain global similarity, etc. Achieve the effect of fast training speed, saving storage overhead, and saving computing overhead

Active Publication Date: 2019-08-30
FUDAN UNIV
View PDF8 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] In order to solve the three problems of low sampling efficiency of existing hash algorithms, failure to maintain global similarity, and hash codes containing redundancy, the present invention proposes a Provides a deep enhanced deredundancy hashing algorithm for image retrieval

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-oriented deep reinforcement redundancy elimination Hash algorithm
  • Image retrieval-oriented deep reinforcement redundancy elimination Hash algorithm
  • Image retrieval-oriented deep reinforcement redundancy elimination Hash algorithm

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0031] It can be seen from the background technology that there are two main defects in the existing hash algorithm for similarity image retrieval. First, most hashing algorithms are trained in small batches, which has low sampling efficiency and cannot maintain global similarity information. Second, the hash codes generated by most hash algorithms contain some redundant bits or even harmful bits. Removing these bits can not only improve the retrieval accuracy, but also reduce computing and storage costs. Therefore, this embodiment addresses the above two problems by using hash code reasoning based on block calculation and hash code de-redundancy based on deep reinforcement learning respectively.

[0032] In this embodiment, the label information is first used to construct a similarity matrix:

[0033]S=min(YY T ,1)×2-1 (6)

[0034] where, S∈{-1,+1} n×n , S ij =+1 means that the i-th image is similar to the j-th image; S ij = -1 means that the i-th image is not similar t...

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 invention belongs to the technical field of image retrieval, and particularly relates to an image retrieval-oriented deep reinforcement redundancy elimination Hash algorithm. The algorithm comprises the steps of carrying out the block hash code reasoning, constructing a similarity matrix by means of the label information of images, then reasoning the optimal hash code of each image according to the similarity matrix, wherein the similarity matrix is huge, and solving in a blocking manner; carrying out the image-hash code mapping, mapping the original pixel information of the image to the reasoned optimal hash code, wherein the mapping process is realized by using multiple classifications; removing the Hash code redundancy bits, removing the Hash bits which are not helpful to retrievalprecision or even harmful to retrieval precision in the generated Hash codes, and during the process, training an agent through the deep reinforcement learning, and finding an optimal mask by the agent, so that the redundant Hash bits can be removed by utilizing the mask. According to the algorithm, the training speed is higher, the calculation cost and the storage cost are saved, and the retrieval precision is high.

Description

technical field [0001] The invention belongs to the technical field of image retrieval, and in particular relates to an image retrieval-oriented depth-enhanced deredundancy hash algorithm. Background technique [0002] With the rapid development of social media, a large amount of multimedia data is generated every day, including text, images, videos, etc. In order to efficiently retrieve these unstructured data, many methods have been proposed. Recently, approximate nearest neighbor retrieval has attracted more and more attention due to its high retrieval accuracy and low computational overhead. Among various approximate nearest neighbor retrieval methods, the hash algorithm is currently the most promising method, which can generate compact binary hash codes for high-dimensional data, and use these hash codes for retrieval in Hamming space . The present invention is concerned with a learning-based hashing algorithm, which is a data-dependent algorithm. Compared with data...

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): G06F16/53G06N5/04G06N3/08
CPCG06F16/53G06N5/04G06N3/08
Inventor 张玥杰杨觉旭张涛
Owner FUDAN UNIV
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