The Analysis of HDR Video Reconstruction Methods

Ghaffari, Mahdi | 2020

163 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 52997 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Amini, Arash
  7. Abstract:
  8. The conventional cameras and displays do not have the ability to record and display the full brightness of the world around us. These deficiencies have led to the development of methods known as High Dynamic Range Imaging. Most of the work done in this field falls into two groups: compression of the light range and its expansion. In compression, the brightness range is intended to display content with a high dynamic range in simple displays. But the goal of expanding the brightness range is to reconstruct HDR content from a low dynamic range one. In addition to the above classification, this field is also categorized based on the type of content (image or video). In this study, the reconstruction(expansion) of HDR content (image or video) has been investigated. There are a variety of ways for HDR image reconstruction compared to HDR videos. The simple way for HDR image reconstruction is combining of several LDR images with different light ranges. But new methods, using deep learning, have used only one LDR image for reconstruction. In this study, it has been tried to simplify the previous neural networks and also to provide more learning images for both the network to be suitable for use in HDR video reconstruction and not to cause a decrease in network performance. The collection of images used in this section is mostly related to HDRIhaven and HDRplus sets. According to the PSNR criterion, the result for the reconstruction of HDR images is 23.6 which is a better result than the previous methods, considering the amount of calculations. The network has 2.5 million parameters, and is simpler than previous methods. The reconstruction of HDR videos will be possible by using HDR image reconstruction methods, and making temporal consistency. For this problem, previous works have been based on the existence of sequences of frames with bracketing exposures, and there have been no similar work to make HDR video with single exposure frames. In this study, HDR videos were reconstructed using the implemented method to reconstruct HDR images and create time consistency for video frames exposure over time. HDRV LiU, Boltard and SJTU frame sequences are used. In this section, using the best proposed method, the temporal error 0.32 has been obtained rather than 0.53 for input sequences based on the warping error for HDR video reconstruction
  9. Keywords:
  10. Deep Learning ; Convolutional Neural Network ; Tone Mapping Operators ; High Dynamic Range Imaging ; High Dynamic Range Video ; Image Reconstruction

 Digital Object List