In this thesis, we primarily focus on one common type of symmetry, the translational symmetry. We first review the current state-of-the-art methods for translational symmetry detection, and discuss their benefits and drawbacks. Towards an efficient, automatic and widely applicable translational symmetry detector, we develop a novel method for automatically detecting translational symmetry patterns, and extracting the corresponding lattices from images without pre-segmentation or reconstructing the underlying 3D geometry. In particular, we employ a region-based feature and fully utilize its regional properties (shape, orientation and well-defined boundary) to propose the repeated candidates. Compared with traditional treatments, which usually rely on point-based features and group them to propose repeated candidates, our treatment is more efficient and stable to perspective projection, distortion or noise. By clustering the candidate regions and indexing the major clusters using a GPU KD-tree, the parallel lattice formation processes turn out to be very efficient and achieve a real-time rate. By using a set of spatially varying vectors with a loose neighboring constraint to represent the underlying lattice, we successfully detect most of translational symmetry patterns over arbitrary surfaces, which can be planar or curve, without or with perspective projection, and even when suffered from global and local deformations. Moreover, the parallel searching and saving scheme enables us to simultaneously detect multiple disjoint symmetry patterns from an input images. / Symmetry has been an important concept in the nature, science and art. There is an abundant of biological, chemical, and artificial structures captured in many real-world images, exhibiting various forms of symmetry. The symmetry patterns and the repetitive elements reinforce the visual importance and usually make an image more attractive. Although our humans have an excellent innate ability in recognizing symmetry and perceiving its beauty, efficient and automatic symmetry detection from images remains a unsolved challenging problem in computer vision and graphics. Without understanding the high-level semantics of symmetry, editing such images while preserving the repetitions and their relations turns out to be difficult to perform, such as image resizing, image inpainting and image replacement. / The significant improvements of our method in both efficiency and accuracy make it a useful tool from which many applications can benefit. One of them is image resizing. We demonstrate that image resizing can be achieved more effectively if we have a better understanding of the image semantics. By analyzing the translational symmetry patterns, and detecting the underlying lattices in an image, we can summarize, instead of only distorting or cropping, the image content. This opens a new space for image resizing that allows us to manipulate, not only image pixels, but also the semantic cells in the lattice. As a general image contains both symmetry & non-symmetry regions and their natures are different, we propose to resize symmetry regions by summarization and non-symmetry region by optimized warping. In addition, by smoothing the intensity of cells across the lattice, we can further maintain the seamlessness of illumination during the summarization. As the difference in resizing strategy between symmetry regions and non-symmetry region leads to discontinuity at their shared boundary, we propose a framework to minimize the artifact. Experimental results show that, with the high-level knowledge of symmetry, our method outperforms the state-of-the-art resizing techniques. / Wu, Huisi. / Advisers: Tien-Tsin Wong; Pheng-Ann Heng. / Source: Dissertation Abstracts International, Volume: 73-06, Section: B, page: . / Thesis (Ph.D.)--Chinese University of Hong Kong, 2011. / Includes bibliographical references (leaves 92-100). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. [Ann Arbor, MI] : ProQuest Information and Learning, [201-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstract also in Chinese.
Identifer | oai:union.ndltd.org:cuhk.edu.hk/oai:cuhk-dr:cuhk_344819 |
Date | January 2011 |
Contributors | Wu, Huisi., Chinese University of Hong Kong Graduate School. Division of Computer Science and Engineering. |
Source Sets | The Chinese University of Hong Kong |
Language | English, Chinese |
Detected Language | English |
Type | Text, theses |
Format | electronic resource, microform, microfiche, 1 online resource (ix, 100 leaves : ill.) |
Rights | Use of this resource is governed by the terms and conditions of the Creative Commons “Attribution-NonCommercial-NoDerivatives 4.0 International” License (http://creativecommons.org/licenses/by-nc-nd/4.0/) |
Page generated in 0.0016 seconds