IP-FSRGAN-CA face image super-resolution reconstruction algorithm based on coordinate attention mechanism

Abstract

In order to solve the problem of feature loss in face feature information extraction and recovery in a traditional face super-resolution image reconstruction technology, the invention provides an IP-FSRGAN-CA face image super-resolution reconstruction algorithm based on a coordinate attention mechanism (CA). Therefore, a generator is helped to learn more face feature information and synthesize a more real super-resolution face picture. When the amplification factor is 4, an LFW data set is adopted to test the model performance, and an experimental result shows that compared with IP-FSRGAN, the method has the advantages that the peak signal-to-noise ratio (PSNR) is improved by 0.14%, the structural similarity (SSIM) is improved by 0.59%, the PSNR on a Y channel is improved by 2.43%, the SSIM is improved by 0.38%, and under the measurement of the PSNR and the SSIM, the method is quantitatively superior to existing SRGAN, ESRGAN, IP-FSRGAN and other human face reconstruction methods. Experiments prove that the IP-FSRGAN-CA face image super-resolution reconstruction algorithm based on the coordinate attention mechanism provided by the invention has effectiveness in face super-resolution image reconstruction.

Description

technical field [0001] The invention belongs to the technical field of super-resolution reconstruction, and in particular relates to a method for super-resolution reconstruction of human face pictures. Background technique [0002] In recent years, generative adversarial networks have been widely used in computer vision, especially in data set expansion, image conversion, text-image conversion, semantic image-image conversion, image editing, image mixing, image restoration, video prediction, 3D printing, etc. It was a huge success. [0003] Single image super-resolution reconstruction is an important research direction in the field of computer vision. For the capture and restoration of face image details, it has very important research significance in the fields of medical imaging, traffic surveillance, criminal investigation and tracking, border guarding, and emergency rescue. In real monitoring scenarios, due to factors such as video collection equipment, natural weather ...

AI Technical Summary

Problems solved by technology

In real monitoring scenarios, due to factors such as video collection equipment, natural weather conditions, unrestricted human activities, and actual project needs, the acquired images may have low resolution, blur or even distortion, which may induce staff to The misjudgment of image information causes immeasurable losses, so the task of face SR image reconstruction is imperative and has a long way to go

Images