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

A Massive Image Classification Method Based on Deep Local Feature Descriptors

A technology of local features and classification methods, applied in the field of image processing and deep learning, to achieve the effect of improving accuracy

Inactive Publication Date: 2018-04-06
UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF4 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, with the sharp increase in the size of image datasets and image diversity, the traditional VLAD gradually shows its inherent disadvantages in massive image classification tasks.
When VLAD is applied to large-scale image sets, the storage of massive high-dimensional VLAD features and subsequent classification operations are extremely challenging for current computer equipment.

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
  • A Massive Image Classification Method Based on Deep Local Feature Descriptors
  • A Massive Image Classification Method Based on Deep Local Feature Descriptors
  • A Massive Image Classification Method Based on Deep Local Feature Descriptors

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0021] The massive image classification method based on the deep local feature descriptor proposed by the present invention specifically includes the following steps:

[0022] Step 1: Set the entire deep learning model as an L-layer deep learning process; extract the SIFT features of each training picture, and the SIFT feature set S of the entire training sample set is expressed as:

[0023] S=[s 1 ,...,s N ], S ∈ R D×N

[0024] Among them, N is the number of SIFT features in the SIFT feature set, and D is the dimension of each SIFT feature; Kmeans clustering is performed on the SIFT feature set S to obtain the dictionary of the first layer of deep learning process D. 1 ∈ R D×K , where K is the dictionary D 1 The number of cluster centers in D, D is the dimension of the cluster centers;

[0025] Step 2: Define that each training picture is a set B=[B of T=M1×M2×M3 image blocks 1 ,...,B t ,...,B T ], that is, each training picture has M3 sub-regions, and each sub-reg...

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 proposes a massive image classification method based on deep local feature descriptors. After extracting multi-level image local feature descriptors, a more abstract and rich image expression is finally obtained, so as to achieve the purpose of improving the accuracy of image classification. The depth model adopted in the present invention can capture more abstract and richer feature information in the image in the layer-by-layer feature extraction, plus the average pooling operation, can make the obtained image features have translation invariance, which is for the increasingly severe The increased number of pictures and the variety of pictures play a key role.

Description

technical field [0001] The invention belongs to the technical field of image processing and deep learning, and relates to high-efficiency massive image processing, in particular to a massive image classification method based on deep local feature descriptors. Background technique [0002] In recent years, image processing has been widely concerned and applied in the fields of industry, manufacturing, military, and medical treatment. Although its development situation is very good, but as the coverage of practical applications gradually widens, massive image data will follow, which makes the traditional image processing methods overwhelmed. The image classification task mainly consists of three parts: image preprocessing, image feature extraction, and classifier selection. Among them, image feature extraction plays a vital role in image classification tasks. [0003] The local feature descriptor (VLAD: vector of locally aggregated descriptors) proposed by Herve Jegou et al....

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): G06K9/66
CPCG06V10/462G06F18/2411
Inventor 董乐
Owner UNIV OF ELECTRONICS SCI & TECH OF CHINA
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