Computer Graphics & Visual Computing (CGVC) 2018

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/2632659

Browse

Browsing Computer Graphics & Visual Computing (CGVC) 2018 by Subject "Image processing"

Now showing 1 - 4 of 4
  • No Thumbnail Available
    Item
    A Deep Learning Approach to No-Reference Image Quality Assessment For Monte Carlo Rendered Images
    (The Eurographics Association, 2018) Whittle, Joss; Jones, Mark W.; {Tam, Gary K. L. and Vidal, Franck
    In Full-Reference Image Quality Assessment (FR-IQA) images are compared with ground truth images that are known to be of high visual quality. These metrics are utilized in order to rank algorithms under test on their image quality performance. Throughout the progress of Monte Carlo rendering processes we often wish to determine whether images being rendered are of sufficient visual quality, without the availability of a ground truth image. In such cases FR-IQA metrics are not applicable and we instead must utilise No-Reference Image Quality Assessment (NR-IQA) measures to make predictions about the perceived quality of unconverged images. In this work we propose a deep learning approach to NR-IQA, trained specifically on noise from Monte Carlo rendering processes, which significantly outperforms existing NR-IQA methods and can produce quality predictions consistent with FR-IQA measures that have access to ground truth images.
  • No Thumbnail Available
    Item
    Groupwise Non-rigid Image Alignment With Graph-based Initialisation
    (The Eurographics Association, 2018) Aal-Yhia, Ahmad; Malcolm, Paul; Akanyeti, Otar; Zwiggelaar, Reyer; Tiddeman, Bernard; {Tam, Gary K. L. and Vidal, Franck
    Groupwise image alignment automatically provides non-rigid registration across a set of images. It has found applications in facial image analysis and medical image analysis by automatically generating statistical models of shape and appearance. The main approaches used previously include iterative and graph-based approaches. In iterative approaches, the registration of each image is iteratively updated to minimise an error measure across the set. Various metrics and optimisation strategies have been proposed to achieve this. Graph-based methods perform registration of each pair of images in the set, to form a weighted graph of the ''distance'' between all the images, and then finds the optimal paths between the most central image and every other image. In this paper, we use a graph-based approach to perform initialisation, which is then refined with an iterative approach. Pairwise registration is performed using demons registration, then shortest paths identified in the resulting graph are used to provide an initial warp for each image by concatenating warps along the path. The warps are refined using an iterative Levenberg-Marquardt minimisation to the mean, based on updating the locations of a small number of points and incorporating a stiffness constraint. This optimisation approach is efficient, has very few free parameters to tune and we show how to tune the few remaining parameters. We compare the combined approach to both the iterative and graph-based approaches used independently. Results demonstrate that the combined method improves the alignment of various datasets, including two face datasets and a difficult medical dataset of prostate MRI images.
  • No Thumbnail Available
    Item
    Image Inpainting for High-Resolution Textures using CNN Texture Synthesis
    (The Eurographics Association, 2018) Laube, Pascal; Grunwald, Michael; Franz, Matthias O.; Umlauf, Georg; {Tam, Gary K. L. and Vidal, Franck
    Deep neural networks have been successfully applied to problems such as image segmentation, image super-resolution, coloration and image inpainting. In this work we propose the use of convolutional neural networks (CNN) for image inpainting of large regions in high-resolution textures. Due to limited computational resources processing high-resolution images with neural networks is still an open problem. Existing methods separate inpainting of global structure and the transfer of details, which leads to blurry results and loss of global coherence in the detail transfer step. Based on advances in texture synthesis using CNNs we propose patch-based image inpainting by a CNN that is able to optimize for global as well as detail texture statistics. Our method is capable of filling large inpainting regions, oftentimes exceeding the quality of comparable methods for high-resolution images. For reference patch look-up we propose to use the same summary statistics that are used in the inpainting process.
  • No Thumbnail Available
    Item
    Segmenting Teeth from Volumetric CT Data with a Hierarchical CNN-based Approach
    (The Eurographics Association, 2018) Macho, Philipp Marten; Kurz, Nadja; Ulges, Adrian; Brylka, Robert; Gietzen, Thomas; Schwanecke, Ulrich; {Tam, Gary K. L. and Vidal, Franck
    This paper addresses the automatic segmentation of teeth in volumetric Computed Tomography (CT) scans of the human skull. Our approach is based on a convolutional neural network employing 3D volumetric convolutions. To tackle data scale issues, we apply a hierarchical coarse-to fine approach combining two CNNs, one for low-resolution detection and one for highresolution refinement. In quantitative experiments on 40 CT scans with manually acquired ground truth, we demonstrate that our approach displays remarkable robustness across different patients and device vendors. Furthermore, our hierarchical extension outperforms a single-scale segmentation, and network size can be reduced compared to previous architectures without loss of accuracy.