Unlock instant, AI-driven research and patent intelligence for your innovation.

Super-resolution method for aligning face images based on residual back projection neural network

A face image and neural network technology, applied in the field of super-resolution, can solve the problems of difficult application of face analysis system and poor visual effect of low-resolution face images, and achieve good visualization effect and high-resolution face images The effect of high structural similarity and high peak signal-to-noise ratio

Inactive Publication Date: 2020-04-10
BEIJING INSTITUTE OF TECHNOLOGYGY
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to address the poor visual effect of existing aligned low-resolution face images, which are difficult to apply to existing face analysis systems. In order to enlarge ultra-low-resolution face images, a neural network alignment based on residual back projection is proposed Super-resolution methods for face images

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
  • Super-resolution method for aligning face images based on residual back projection neural network
  • Super-resolution method for aligning face images based on residual back projection neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0033] This example illustrates the specific implementation of the super-resolution method for aligning face images based on the residual back-projection neural network of the present invention.

[0034] When implementing the super-resolution method for aligning face images of the present invention, an open-source face image data set celebA data set is used for testing. The data set contains a total of 200,000 frontal face images. We randomly sampled 5000 face images as a validation set, 1000 face images as a test set, and the rest as a validation set. Except for the different data sets used, the training, verification and testing steps of the neural network are the same. The experimental environment adopted by the present invention: the hardware system is a TiTan X independent graphics card, the video memory is 12G, the software system is ubuntu14.04, and the pythonpytorch framework is used. Using peak signal-to-noise ratio (PSNR), and structural similarity measure (SSIM) as...

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 relates to a super-resolution method for aligning face images based on a residual back projection neural network and belongs to the technical field of image processing. According to themethod, an iterative back projection and deep learning neural network combined mode is adopted, and an ultralow-resolution face image is amplified by 8 times through the following three steps of: (1)inputting an ultralow-resolution face image into a neural network, extracting depth features, and amplifying a low-resolution feature image to a size of 128*128 by adopting a deconvolution network; (2) inputting the 128*128 feature image obtained in the step (1) into the residual back projection unit of the neural network, and a compensated 128*128 high-resolution feature image is obtained throughcontinuous iteration; and (3) generating a final 128*128 high-resolution image from the high-resolution feature image obtained in the step (2) through a convolution layer. The method is clear in module and simple in step. The super-resolution effect and efficiency of the method meet the super-resolution requirement of an actual low-resolution face image.

Description

technical field [0001] The invention relates to a super-resolution method for aligning face images based on a residual back-projection neural network, and belongs to the technical field of image processing. [0002] technical background [0003] In the research field of computer vision, face image super-resolution has always been an important sub-topic. It not only has many practical application scenarios, but also is the basis of other research topics. [0004] From a practical point of view, many intelligent applications are inseparable from the support of face image super-resolution technology. The most important application is the urban surveillance system: with the rapid economic development, the video surveillance cameras around us are becoming more Many, these cameras are mainly used to build urban video surveillance systems, and play an important role in the criminal investigation business of public security organs. However, in the actual process of collecting faces ...

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): G06K9/00G06T3/40
CPCG06T3/4053G06V40/161G06V40/168
Inventor 陆耀王学博陈晓珍王子建李玮琪李公平吴紫薇
Owner BEIJING INSTITUTE OF TECHNOLOGYGY