Advanced

Deep Learning Methods for Streaming Image Reconstruction in Fixed-camera Settings

Förberg, Johan LU (2017) In Master's Theses in Mathematical Sciences FMAM05 20172
Mathematics (Faculty of Engineering)
Abstract
A streaming video reconstruction system is described and implemented as a convolutional neural network. The system performs combined 2x super-resolution and H.264 artefacts removal with a processing speed of about 6 frames per second at 1920×1080 output resolution on current workstation-grade hardware. In 4x super-resolution mode, the system can output 3840×2160 video at a similar rate. The base system provides quality improvements of 0.010–0.025 SSIM over Lanczos filtering. Scene-specific training, in which the system automatically adapts to the current scene viewed by the camera, is shown to achieve up to 0.030 SSIM additional improvement in some scenarios. It is further shown that scene-specific training can provide some improvement... (More)
A streaming video reconstruction system is described and implemented as a convolutional neural network. The system performs combined 2x super-resolution and H.264 artefacts removal with a processing speed of about 6 frames per second at 1920×1080 output resolution on current workstation-grade hardware. In 4x super-resolution mode, the system can output 3840×2160 video at a similar rate. The base system provides quality improvements of 0.010–0.025 SSIM over Lanczos filtering. Scene-specific training, in which the system automatically adapts to the current scene viewed by the camera, is shown to achieve up to 0.030 SSIM additional improvement in some scenarios. It is further shown that scene-specific training can provide some improvement even when reconstructing an unfamiliar scene, as long as the camera and capture settings remain the same. (Less)
Popular Abstract (Swedish)
Många kameror sitter fast monterade och filmar samma plats varje dag. Tänk om kameror kunde tränas till att minnas vad de sett?
Please use this url to cite or link to this publication:
author
Förberg, Johan LU
supervisor
organization
course
FMAM05 20172
year
type
H2 - Master's Degree (Two Years)
subject
keywords
image reconstruction, image enhancement, super-resolution, artefacts removal, deep learning, machine learning, convolutional neural networks, computer vision
publication/series
Master's Theses in Mathematical Sciences
report number
LUTFMA-3336-2017
ISSN
1404-6342
other publication id
2017:E68
language
English
id
8929567
date added to LUP
2018-06-07 17:45:42
date last changed
2018-06-07 17:45:42
@misc{8929567,
  abstract     = {A streaming video reconstruction system is described and implemented as a convolutional neural network. The system performs combined 2x super-resolution and H.264 artefacts removal with a processing speed of about 6 frames per second at 1920×1080 output resolution on current workstation-grade hardware. In 4x super-resolution mode, the system can output 3840×2160 video at a similar rate. The base system provides quality improvements of 0.010–0.025 SSIM over Lanczos filtering. Scene-specific training, in which the system automatically adapts to the current scene viewed by the camera, is shown to achieve up to 0.030 SSIM additional improvement in some scenarios. It is further shown that scene-specific training can provide some improvement even when reconstructing an unfamiliar scene, as long as the camera and capture settings remain the same.},
  author       = {Förberg, Johan},
  issn         = {1404-6342},
  keyword      = {image reconstruction,image enhancement,super-resolution,artefacts removal,deep learning,machine learning,convolutional neural networks,computer vision},
  language     = {eng},
  note         = {Student Paper},
  series       = {Master's Theses in Mathematical Sciences},
  title        = {Deep Learning Methods for Streaming Image Reconstruction in Fixed-camera Settings},
  year         = {2017},
}