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

Illumination Domain Normalization Method of Color Face Image Based on Recurrent Generative Adversarial Network

A color face and face image technology, applied in the field of computer vision face lighting, can solve problems such as limiting the adaptability of face images, poor lighting normalization effect, and application limitations

Active Publication Date: 2021-03-26
SUN YAT SEN UNIV
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

This limits the adaptability of the method to face images outside the training dataset. For cross-dataset testing, the normalization effect of the generated illumination is not good, and its application also has great limitations.

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
  • Illumination Domain Normalization Method of Color Face Image Based on Recurrent Generative Adversarial Network
  • Illumination Domain Normalization Method of Color Face Image Based on Recurrent Generative Adversarial Network
  • Illumination Domain Normalization Method of Color Face Image Based on Recurrent Generative Adversarial Network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0077] The present invention is aimed at the normalized conversion of multi-illumination domains of face color pictures. The data set we use is the face data set of CMU-multi-PIE, which is collected by the visual research team of Carnegie Mellon University in the United States. The data The set of images contains 337 face identities, and each face identity has 20 pictures captured under 19 different lighting conditions. The first and last pictures are not illuminated by any camera flash, and the rest are individually illuminated by 18 ring flashes. Flash shot. We selected 18,420 frontal face images with neutral expressions, among which 1,800 were randomly selected as the test set, and the remaining 16,620 were used as the training set, and the input and output were adjusted to 128x128. From the training set, uniformly illuminated face pictures with illumination categories of 06, 07, and 08 are selected to construct the target illumination inter-domain set B, and the remaining ...

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 color face image illumination domain normalization method based on a cyclic generative adversarial network. The method comprises the following steps of S1, establishing a cyclic generative adversarial network model for color face image illumination normalization; S2, establishing a loss function of the model; and S3, carrying out model training, and testing on the test set. According to the invention, conversion to a specified target illumination domain is carried out on color face images under various illumination conditions; a color face image with non-uniform illumination is inputted, a cyclic generative adversarial network is used as a model architecture, a target uniform illumination domain is used as a target, multi-illumination normalization of the face image is realized, and the normalized image not only can well maintain the face attribute characteristics of an original face, but also can well realize the cross-data-set migration.

Description

technical field [0001] The present invention relates to the field of computer vision face illumination, and more specifically, to a method for normalizing the illumination domain of color face images based on recurrent generative adversarial networks. Background technique [0002] In recent years, due to the rapid development of deep learning in computer vision, analysis technologies based on two-dimensional faces, such as face recognition, face matching, and face attribute recognition, have received great attention. At present, many algorithms in the face field can achieve near-perfect performance and have been widely used. These algorithms with good performance are based on the strict control of non-face identity information such as the pose of the face image, shooting environment, expression, and lighting. For the natural and unrestricted real environment, its application will exist Certain deficiencies. Under natural conditions, its performance is easily affected by ma...

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
IPC IPC(8): G06K9/00G06N3/04
Inventor 朱俊勇李锴莹赖剑煌谢晓华
Owner SUN YAT SEN UNIV