Browsing by Author "Lin, Mengchen"

Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemOpen Access
    A Novel Deep CNN-based Tone Mapping Operator and Wide Dynamic Range Imaging Applications
    (2020-12) Lin, Mengchen; Jacobson, Michael John; Yadid-Pecht, Orly; Dimitrov, Vassil S.; Safavi-Naini, Reihaneh S.
    This thesis addresses three Wide Dynamic Range (WDR) imaging and application problems. First, we present a novel tone mapping solution using a reformulated Laplacian pyramid and deep learning. The reformulated Laplacian pyramid constantly decomposes a WDR image into two frequency bands where the low-frequency band is global feature-oriented, and the high-frequency band is local feature-oriented. The reformulation preserves the local features in their original resolution and condenses the global features into a low-resolution image. The generated frequency bands are reconstructed and fine-tuned to output the final tone mapped image that can reveal more detail and contrast. Then we discuss the ability of the model to recover and enhance the lost detail of a degraded. We adopt the same theory to build a deep single image enhancer. Our model outperforms the state-of-the-art methods. Next, we propose a WDR dataset for face detection, named WDR FACE, to facilitate and support future face detection research in the WDR field. This dataset contains a total of 398 16-bit megapixel grayscale WDR images collected from 29 subjects and eight selected WDR scenes. The dynamic range of 90\% images surpasses 60,000:1. We also provide the preliminary experimental results of face detection with 25 different tone mapping operators and five different face detectors on this WDR dataset. We provide preliminary results of face detection experiments as a reference for researchers using this data set in the future. Finally, we applied WDR imaging techniques to Magnetic Resonance (MR) imaging to provide better visualization of the images to physicians. We carried out a feasibility study using domain expert observation to show that it is possible to assist physicians' diagnosis of knee lesion injuries via MR image contrast enhancement using different tone mapping algorithms.