Method and apparatus for extracting hash code from image, and image retrieval method and apparatus

An image, redundant hash code technology, applied in still image data retrieval, image coding, image data processing, etc. The effect of reducing information redundancy, simplifying network structure, and improving accuracy

Active Publication Date: 2019-01-04
BEIJING QIHOO TECH CO LTD
View PDF5 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The variational pruning of the VAE framework will cause some hidden layer units to collapse when they are not effectively extracted in the early stage of model training, so that the framework has obvious inherent deficiencies, for example, (1) There are many coding spaces Redundant dimension (i.e., redundant data without information); (2) The framework underutilizes the latent code of the coding space; etc.
Especially when the decoder structure is complex, these shortcomings are more obvious
This will lead to: the inability to accurately extract the image hash code, resulting in a decrease in the accuracy of image retrieval and other related application problems

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
  • Method and apparatus for extracting hash code from image, and image retrieval method and apparatus
  • Method and apparatus for extracting hash code from image, and image retrieval method and apparatus
  • Method and apparatus for extracting hash code from image, and image retrieval method and apparatus

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0153] Embodiment 1, MNIST data and image reconstruction

[0154] Network parameter settings: the number of encoder and decoder layers M is set to 1, δ and η are 0.01, and the prior parameters ρ j is 0.5, the threshold parameter ∈ is set to 0.05, the encoder input data and decoder output data dimensions are 28*28=784, the hash code and the hidden layer data dimensions of the encoder and decoder are 64;

[0155] The training data set is the training set of MIST; the batch size of data samples used in each step of training is set to 32; the image reconstruction error of the model is evaluated at different numbers of training rounds, and the evaluation uses the MIST test set, which is extracted by the encoder. The code and decoder generate reconstructed data, and calculate the error between input and reconstructed data. The calculation method is as follows:

[0156]

[0157] Where N is the number of evaluation samples, D is the dimensionality of each sample data, x is the inp...

Embodiment 2

[0159] Embodiment 2, CIFAR-10 image retrieval

[0160] Network parameter settings: the number of encoder and decoder layers M is set to 4, δ is 0.01, η is 0.01, and the prior parameter ρ j is 0.5, the threshold parameter ∈ is set to 0.05, the dimension of encoder input data and decoder output data is 512, and the dimensions of hash code and hidden layer data of encoder and decoder are 32, 64 and 128;

[0161] 100 sample data are randomly selected from each of the 10 types of data in the CIFAR-10 dataset, a total of 1000 sample data are used as the retrieval input during testing, and the rest of the data are training samples and image databases. For each step of training, the batch size of data samples is set to 32, and the number of training rounds is 200.

[0162] Use the mAP index to evaluate the image retrieval ability. The mAP results of the three models with hash code dimensions of 32, 64 and 128 are shown in Table 1. Table 1 shows the mAP (%) test results of SGH and R-S...

Embodiment 3

[0166] Embodiment 3, Caltech-256 image retrieval

[0167] Network parameter settings: the number of encoder and decoder layers M is set to 4, δ is 0.01, η is 0.01, and the prior parameter ρ j is 0.5, the threshold parameter ∈ is set to 0.05, the dimension of encoder input data and decoder output data is 512, and the dimensions of hash code and hidden layer data of encoder and decoder are 32, 64 and 128;

[0168] 1000 sample data are randomly selected from the Caltech-256 dataset as the retrieval input during testing, and the rest of the data are training samples and image libraries. For each step of training, the batch size of data samples is set to 32, and the number of training rounds is 200.

[0169] Use the mAP index to evaluate the image retrieval ability. The mAP results of the three models with hash code dimensions of 32, 64 and 128 are shown in Table 2. Table 2 shows the mAP (%) of SGH and R-SGH in the Cal tech-256 data set Test Results:

[0170] Table 2 mAP(%) test...

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 a method, a device and an image retrieval method for extracting a hash code from an image. The device comprises the following steps: a hash code extraction model being constructed, wherein the model comprises an encoder and a decoder, and the encoder is composed of a multi-layer depth neural network DNN; extracting hash code from image data and outputting it to decoder, wherein the decoder is composed of multi-layer DNN, which converts the input hash code into image; regularizing the output of the last layer of the decoder so as to ensure that the output of the hidden layer of DNN is close to the hash code as far as possible, thereby simplifying the network structure of the decoder and forcing the encoder to extract high-quality hash code to obtain an anti-redundanthash code depth extraction model; training the depth extraction model of anti-redundant hash codes to determine the parameters in the model; the hash code being extracted from the image by using thetrained anti-redundant hash code depth extraction model encoder. The method can effectively reduce the redundancy of the encoded spatial information and extract the image hash code with high precisionby using all dimensions.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a method and device for extracting a hash code from an image, an image retrieval method and device, electronic equipment and a computer-readable storage medium. Background technique [0002] LTH (learning to hash) is an image compression method that is very effective in image retrieval applications. This framework extracts binary hash codes from images, calculates the similarity between the input image and the image hash codes in the image library, and performs retrieval. The LTH framework can greatly reduce storage space and improve retrieval efficiency. [0003] The hash code extraction of the image in LTH is very critical, and it is generally implemented by an encoder. An autoencoder is an unsupervised neural network method consisting of an encoder and a decoder that generates images from a random code. VAE (Variational Autoencoder, Variational Autoen...

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): G06F16/53G06T9/00G06N3/04
CPCG06T9/00G06N3/045
Inventor 王浩杜长营庞旭林张晨杨康
Owner BEIJING QIHOO 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