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

An Image Description Method Based on Convolutional Recurrent Mixture Model

A technology of cyclic mixing and image description, applied in neural learning methods, biological neural network models, still image data retrieval, etc., can solve problems such as inability to describe content, and achieve the effect of improving application capabilities

Inactive Publication Date: 2019-06-14
BEIJING UNIV OF TECH
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] There have been many methods for image description. However, these models often rely on some hard-coded visual concepts and some regularized templates. fully describe

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
  • An Image Description Method Based on Convolutional Recurrent Mixture Model
  • An Image Description Method Based on Convolutional Recurrent Mixture Model
  • An Image Description Method Based on Convolutional Recurrent Mixture Model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0055] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation examples.

[0056] Flowchart of image description methods applied in machine vision and natural language processing. Such as figure 1 shown.

[0057] It is characterized in that it comprises the following steps:

[0058] Step 1, encode the image, the specific steps are as follows:

[0059] Step 1.1, feature extraction is carried out to image with convolutional neural network, adopted VGG network structure, this network carries out parameter learning on ImageNet data set; Input a training image I t , through the network for feature extraction, and finally get a feature vector F with a size of 4096 t ;

[0060] Step 1.2, through a 4096*256 mapping matrix W e For the extracted feature vector F t Encoding is performed, and a vector v of size 256 is obtained after encoding:

[0061] v=F t T W e +b m (1)

[0062] where W e is a mapping...

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 present invention discloses an image description method based on a convolution cyclic hybrid model, and belongs to the deep learning field of machine learning. For text description, due to the fact that words in sentences have strong context relationships, text data can be encoded by using a language model. The image description method specifically comprises the steps of (1) extracting image characteristics; (2) encoding the image characteristics; (3) encoding image description texts; (4) training the model; and (5) generating image text description by utilization of the trained model. The image description method is widely used in machine vision and natural language processing, and new thought and solutions are provided in an image description method aspect. At present, in image description, text encoding is randomly generated, which has a certain blindness, and the effect is not good. The texts are encoded by utilization of word2Vec, the encoding problem of the description texts in image description is solved, and the defects of randomness, blindness and instability are remedied. Application ability of the image description is largely increased, and foundation is established for development of machine vision.

Description

technical field [0001] The invention belongs to the deep learning part in machine learning. The specific content is the methods applied in the fields of computer vision, natural language processing and image description. Background technique [0002] With the growth of the scale of the Internet and digital information resources, the amount of information has increased exponentially, and the field of information services is facing the dilemma of "rich information, but difficult to obtain useful information". Especially since the 21st century, digital image resources have grown exponentially, causing users to encounter great difficulties in image retrieval in massive image databases, making useful images unable to be corrected within an effective time in massive images. retrieved. In ILSVRC2012, AlexKrizhevsky proposed a five-layer neural network referred to as AlexNet. This network is very complex with 60 million parameters. Finally, AlexNet won the first place in the compe...

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 Patents(China)
IPC IPC(8): G06F16/58G06N3/08
CPCG06F16/5866G06N3/084
Inventor 李玉鑑丁勇刘兆英
Owner BEIJING UNIV OF TECH
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