Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Image data rapid retrieval method based on Hash learning

A technology of image data and hashing, applied in the field of fast retrieval of image data, can solve problems such as model deviation, and achieve the effect of good superiority, good visual effect and good performance

Inactive Publication Date: 2019-08-16
HARBIN ENG UNIV
View PDF16 Cites 10 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] In order to solve the problem that the existing model uses multiple relaxations in the hash code generation stage, the model will have deviations in the negative feedback process of the training stage, and provides a fast retrieval method for image data based on hash learning

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 data rapid retrieval method based on Hash learning
  • Image data rapid retrieval method based on Hash learning
  • Image data rapid retrieval method based on Hash learning

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment approach 1

[0041] The image data fast retrieval method based on hash learning of the present embodiment specifically includes the following steps:

[0042] 1. Establish a deep hash model:

[0043] The deep hash model consists of five convolution-pooling layers, two fully connected layers, a feature layer, a hash layer, and an output layer; the feature layer outputs a feature vector of a certain length; then the feature vector is mapped to the hash through the hash layer code. The model structure is as figure 1 Shown, the specific parameters are shown in Table 1.

[0044] Table 1 Model parameters

[0045]

[0046] Structurally, each fully connected layer consists of a single layer of 500×1 neurons and an activation function. The role of the fully connected layer is to connect each feature of the intermediate feature layer, and the relationship between the extracted features corresponds to the different bits of the hash code through the hash layer, so that different sample images ca...

specific Embodiment approach 2

[0067] The process of using the hash code of the query image and the image sample library for retrieval described in this embodiment includes the following steps:

[0068] The hash function obtained by the deep hash model can make each sample image in the sample library have a unique hash code {h 1 ,h 2 ,..., h m}, h i ∈{0,1}. When it is necessary to retrieve similar images of the query sample q, the formula for calculating the Hamming distance from the images in the sample library is:

[0069]

[0070] In the formula, dist H (h i , h j ) is the Hamming distance, and m is the length of the hash code. It can be known from the formula that each bit in the hash code has the same function, and in the process of generating the hash code, each bit of the hash code is a single feature or a combination of multiple features. When using the Hamming distance Ignored when searching. In addition to being unable to express features, in the image retrieval, the retrieval results ...

Embodiment

[0086] Using CIFAR-10 (A. Krizhevsky, G. Hinton. Learning Multiple Layers of Features from Tiny Images [J]. 2012.) and NUS-WIDE (Zhang P, Zhang W, Li W J, etal. Supervised hashing with latent factor models [ M].2014.) data set for experiments to ensure the effectiveness and reliability of experimental comparisons. The experiment extracts 600 image samples from each category in the CIFAR-10 dataset as experimental data, of which 500 image samples are used as training data, and the other 100 image samples are used as test data. Since the NUS-WIDE dataset is a multi-label dataset, if two sample images have the same label, they are considered to be the same sample data. In the experiment, using the same calculation method as others, take the average mAP of the first 5000 returned samples as the final comparison data. It can be seen from the results that FastH, CNNH, and NINH combined with deep neural networks have better accuracy than traditional methods. In CNNH, the hash code ...

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 discloses an image data rapid retrieval method based on Hash learning, relates to an image data rapid retrieval method, and belongs to the technical field of data retrieval. The problemthat the negative feedback process of an existing model deviates in the training stage due to the fact that the existing model is used for multiple times of relaxation in the Hash code generation stage is solved. A deep hash model contains five convolution-pooling layers, two fully connected layers, feature layers, hash layers, and output layers. The method comprises the following steps that training is carried out based on the triplet constraint, after the trained deep hash model is acquired, the sample library is constructed by using the deep hash model; the sample library is composed of image samples and corresponding hash codes; for the query image, the trained depth is used. The hash model generates a hash code of the query image; the hash code of the query image and the image samplelibrary are used for retrieval.. The method is suitable for image data retrieval.

Description

technical field [0001] The invention relates to a fast retrieval method for image data, belonging to the technical field of data retrieval. Background technique [0002] With the rapid development of the Internet in recent years, high-dimensional data has shown an exponential development. How to use these data has become the focus of various industries. In the past, researchers have proposed many methods for large-scale data retrieval. Hash methods are widely used due to their high storage and computing efficiency (LI WuJun, ZHOU ZhiHua. Big Data Hash Learning: Status and Trends[J] . Science Bulletin, 2015, 60(Z1): 485-490.). Traditional hashing methods, including locality-sensitive hashing and spectral hashing, have achieved certain results in image retrieval, but there is still a distance from practical applications. The rapid development of deep learning promotes the progress of the hashing method. In 2014, Pan Yan and Yan Shuicheng combined the convolutional neural net...

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): G06F16/51G06F16/583
CPCG06F16/51G06F16/583
Inventor 王红滨纪斯佳张毅周连科王念滨童鹏鹏崔琎
Owner HARBIN ENG UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com