Projects

H    SOLUTIONS    Projects

PROJECTS

Development of deblurring algorithm
  • Development of deblurring algorithm 1번 상세이미지 썸네일

Development of deblurring algorithm

Development of deblurring algorithm

Development background

In the CCTV environment, face recognition rate can be lowered due to image quality degradation

→ Image blurred due to movement and compression artifacts

Project details

A function that supplements the recognition rate by improving image quality as object detection/classification and face recognition rates may be lowered due to image quality degradation in the CCTV environment

Details of deblurring algorithm development

deblurring 01.png
  • 1. Development of deblurring algorithm to remove blur caused by motion
  • 2. Video deblurring model established based on deep learning
  • 3. Deblurring datasets created for training and evaluation

Performance of deblurring algorithm

- Improved performance and speed with advanced training and inference technique

Current performance
Method PSNR SSIM FPS
IFI_RNN (C1H1) 28.79 0.8647 57.93
IFI_RNN (C1H2) 29.03 0.8712 45.53
IFI_RNN (C1H3) 29.07 0.8730 37.16
IFI_RNN (C1H4) 29.16 0.8730 30.67
IFI_RNN (C2H1) 29.72 0.8884 41.50
IFI_RNN (C2H2) 29.72 0.8884 34.04
IFI_RNN (C2H3) 29.97 0.8947 28.84
IFI_RNN (C2H4) 29.93 0.8943 25.29
Improved performance
PSNR SSIM FPS
29.05 (+0.26) 0.8749 (+0.011) 111.93 (x1.93)
29.59 (+0.56) 0.8887 (+0.018) 81.96 (x1.80)
29.81 (+0.74) 0.8943 (+0.021) 68.78 (x1.85)
29.83 (+0.67) 0.8950 (+0.022) 56.81 (x1.85)
30.28 (+0.56) 0.9037 (+0.015) 65.05 (x1.56)
30.73 (+1.01) 0.9125 (+0.024) 53.22 (x1.56)
30.72 (+0.75) 0.9121 (+0.017) 46.75 (x1.62)
30.76 (+0.83) 0.9126 (+0.018) 41.42 (x1.63)

Results of deblurring algorithm development

deblurring 03.png