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

Low-resolution image recognition algorithm for compensating edge information

A low-resolution image and high-resolution image technology, applied in the field of low-resolution image recognition algorithms, can solve the problems of large convolution kernel size, unsatisfactory recognition effect, and inability to effectively extract various features, so as to improve the recognition rate , Recognition algorithm robust effect

Pending Publication Date: 2021-05-07
XIAN UNIV OF POSTS & TELECOMM
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The size of the convolution kernel of the classic neural network recognition model is often relatively large, while the size of the low-resolution image is small. When the feature is extracted through the convolution operation, various features cannot be effectively extracted, so the recognition effect is not ideal.

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
  • Low-resolution image recognition algorithm for compensating edge information
  • Low-resolution image recognition algorithm for compensating edge information
  • Low-resolution image recognition algorithm for compensating edge information

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0040] The image recognition module of the present invention is mainly used for recognizing low-resolution images. This module uses the classic LeNet-5 model, which is an end-to-end network with a total of 7 layers, including 2 convolutional layers, 2 pooling layers and 3 fully connected layers. The parameters of each layer are shown in Table 1. . Traditional recognition networks directly convert low-resolution images to sent to the network for training, and the present invention is different from the traditional strategy, which uses low-resolution images and its estimated edge information image C pred After fusion, the input image is sent to the network for training, and the edge information image enhances the high-frequency detail information in the low-resolution image, so the recognition network can capture richer image features, thereby improving the recognition rate of low-resolution images .

[0041] Table 1 The parameters of each layer of LeNet-5 recognition netw...

Embodiment 2

[0044] The edge information generation module of the present invention is mainly used for estimating the edge information of the high-resolution image based on the low-resolution image. This module adopts the structure of generative confrontation network, which is divided into two parts: generation network and discriminant network. The specific network structure parameters are shown in Table 2.

[0045] Table 2 Network parameters of each layer of edge information generation module

[0046]

[0047]

[0048] The specific generation steps are as follows:

[0049] step1: input image preparation. The original high-resolution images in the training dataset I gt N times downsampling, and then N times upsampling to generate a low resolution image with the same size as the high resolution image Then use the Canny operator to extract the edge information of the high-resolution image and the low-resolution image respectively, and obtain the edge C of the high-resolution image...

Embodiment 3

[0057] The residual module containing the attention mechanism of the present invention is mainly used to select the features that are more critical to the prediction edge task from many features, and at the same time strengthen the anti-noise performance of the generation network. The specific network structure and its parameters are shown in Table 2.

[0058] Step1: Generate channel attention feature F1. The output features of the down-sampling module are sent to the feature extraction module to obtain the feature F1, and then the features are sent to the multi-layer perceptron with a hidden layer after passing through the maximum pooling layer and the average pooling layer, and then the The results are summed to obtain the channel attention feature F′.

[0059] step2: Generate the first correction feature F2. The channel attention feature F' is multiplied with the output feature of the feature extraction module to obtain the first modified feature F2.

[0060] step3: Gene...

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 low-resolution image recognition algorithm for compensating edge information. The algorithm comprises: an image recognition module which used for recognizing a low-resolution image; and an edge information generation module which is used for estimating edge information approximate to the high-resolution image according to the low-resolution image. According to the low-resolution image recognition algorithm, the edge information of the low-resolution image is predicted by using the low-resolution image, the edge information contains more lost high-frequency detail information in the low-resolution image, and the identification rate of the low-resolution image can be improved by using the predicted edge information. Meanwhile, an attention mechanism module is added into a residual block of the edge generation model, so that the extracted features can be more beneficial to generating the edge of an approximate high-resolution image, and certain anti-noise performance is achieved.

Description

technical field [0001] The invention belongs to the field of image processing, in particular to a low-resolution image recognition algorithm for compensating edge information. Background technique [0002] In recent years, the promotion and popularization of monitoring equipment has brought great convenience to the investigation and solving of crimes. However, due to factors such as illumination, shooting distance, and shooting angle, the resolution of the obtained target image is low, only a dozen to dozens of pixels, and some images still have speckle noise. Therefore, how to identify the target object in the low-resolution image is an urgent practical application problem to be solved. [0003] Image recognition is a classic problem in the field of image processing and pattern recognition. At present, there are very excellent algorithms, such as LeNet-5 model, VGG16 model, DeepFace algorithm, Face++ algorithm, etc., and its image recognition rate can reach 99%. However, ...

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): G06T7/13
CPCG06T7/13G06T2207/20081G06T2207/20084G06T2207/20192
Inventor 毕萍刘玉霞谭仕立刘颖
Owner XIAN UNIV OF POSTS & TELECOMM
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