臺灣博碩士論文加值系統

本篇論文提出根據一系列的影像來建造球形全景影像的技術。此方法是將每一張輸入的影像關連至某一組轉換矩陣，而不採用將所有的影像先行投影至某一共同曲面（例如柱狀面）的傳統作法。特別是當建造一個擁有完整視野的全景影像時，我們引入不含有平移運動的平面運動模型，也就是對每一張輸入的影像，求出其對應的旋轉矩陣（因不含平移運動）以及焦距的長度。為了快速地求得更精確的數值，我們也發展了另一個稱為區域搜尋的方法，以修正偏差較大的估計值。由粗糙到精細的影像階層式架構也被採用以彌補相鄰影像間較大的位移，並同時增進系統的效率。
對於一般使用的環景影像而言，在可允許的俯仰角上仍存在相當大的限制。我們可藉由接合不同緯度的多環來擴展垂直方向的視野，然而這引發了另一個問題：影像的旋轉矩陣無法確定。在同時存在多個可能解的情況下，我們提出了另一個解決不確定性的方法：內插運算。透過對旋轉矩陣的內插運算，可以得到另一個新的內插矩陣。這樣的方法可以減少累積的誤差，在合成的全景影像便不會產生導源於定位錯誤的鬼影現象。
我們同時也實作了一個包含自動與手動接圖功能的整合編輯環境，可視為一個可用來編輯場景的瀏覽系統。不同於一般的三維場景模型所使用的三角片或曲面，此整合編輯環境使用影像面來建造虛擬真實的人造場景。易於使用的設計及操控使得此編輯環境可作為前置處理之用，以提供夠接近的估計值作為自動接圖工具的初始設定。它支援各種型態的三維轉換來控制所有的影像面，以得到一較佳的視覺效果。
除此之外，我們也發展了一套環境映射圖產生的方法。在輸出的格式中，我們採用了傳統經緯映射圖的格式，故可使用一般的全景影像瀏覽器來觀看，如Live Picture 瀏覽器。

This paper proposes some techniques for constructing full view panoramic image mosaics from a sequence of images. Rather than explicitly projecting all of the images onto a common surface (e.g., a cylinder), our mosaic representation associates a transformation matrix with each input image. In particular, to construct a full view panorama, we introduce a planar surface motion model without motion parallax that associates a pure rotational matrix and a focal length with each input image. A refinement approach to the estimated rotational representation called local search is developed to quickly fix an incorrect estimation. To compensate for large displacements between a pair of images, a coarse-to-fine hierarchical structure is used. This strategy also increases the system performance.
To expand the vertical view or even create a spherical panorama without viewing restrictions, we can stitch multiple "image rings" which are located at different latitudes. However, concatenating consecutive rings involves another problem: relationship ambiguity. To solve this problem, we first build the 2D topology of all images. Then according to the topology graph established, a cross-pairwise registration is performed and a quanternion interpolation strategy is adopted to reduce accumulated error and minimize the blurring effects.
An integrated editing environment (IEE) combining both manual and automatic editing function is implemented. The IEE is a designed as a scene construction tool with arbitrary browsing ability. Instead of using triangles or meshes to model a 3D virtual world, our IEE uses different primitives which are called image planes to establish a semi-realistic artificial scene. Friendly design and easy manipulation make the manual editing environment a useful pre-processing tool to provide a close initial guess to auto-registration function, which is critical to a local minimum optimization algorithm similar to what we used before.
In addition, an environment map generation strategy is developed. We use the traditional longitude-latitude projection map as the output format. A specialized panoramic browser such as Live Picture viewer can be used to navigate the panorama.

1 INTRODUCTION3
2 RELATED WORK6
2.1 PANORAMIC HARDWARE6
2.1.1 IPIX7
2.2 CYLINDRICAL PANORAMAS8
2.2.1 PhotoVista10
3 A MANUAL EDITING SYSTEM12
3.1 OVERVIEW12
3.2 BUILD UP THE REAL 3D WORLD13
3.3 VISUALIZATION14
3.3 NEIGHBOR LINKING AND IMAGE MANAGEMENT16
3.4 ELIMINATING INTENSITY DISCONTINUITY19
3.5 PANORAMA GENERATION20
4 TWO KINDS OF IMAGE MOSAICING21
4.1 TRANSFORMATION MODELS21
4.2 INTENSITY BASED MOSAICING24
4.3 FEATURE BASED MOSAICING26
4.3.1 Template matching and feature matching26
4.3.2 Epipolar geometry27
4.3.3 Finding candidate matches29
4.3.4 Estimation of the epipolar geometry31
5 AN AUTOMATIC IMAGE REGISTRATION ALGORITHM33
5.1 PLANAR SURFACE FLOW MOTION MODEL34
5.2 ROTATIONAL MODEL35
5.3 LOCAL SEARCH39
5.4 PYRAMID CONSTRUCTION40
5.5 CLOSING HORIZONTAL GAPS43
5.6 EXPANDING THE VERTICAL VIEW FIELD43
5.7 BLENDING AND CREATING MOSAICS46
5.8 EXPERIMENTAL RESULTS47
6 CONCLUSIONS55
7 FUTURE WORK57
REFERENCES59
APPENDIX Ι61
I.1 PROBLEM DESCRIPTION61
I.2 THE MATHEMATICAL INFERENCE62

[1] Shenchang Eric Chen. QuickTime VR - An Image-Based Approach to Virtual Environment Navigation. Computer Graphics (SIGGRAPH''95), pages 29-38, 1995.
[2] Ju-Wei Chen, Shu-Cheng Huang, Tse Cheng, Der-Lor Way, and Li-Sheng Shen. Building Full View Spherical Panoramas via an Image-Based Approach. International Computer Symposium Workshop on Computer Graphics and Virtual Reality, pages 112-119, 1998.
[3] Richard Szeliski and Heung Yeung Shum. Creating Full View Panoramic Image Mosaics and Environment Maps. In Computer Graphics Proceedings, Annual Conference Series, pages 251-258, Proc. SIGGRAPH''97, 1997.
[4] James R. Bergen, P. Anandan, Keith J. Hanna, and Rajesh Hingorani. Hierarchical Model-Based Motion Estimation. In Second European Conference on Computer Vision (ECCV''92), pages 237-252, Santa Margherita Liguere, Italy, May 1992.
[5] S. Mann and R. W. Picard. ''Video orbits'': characterizing the coordinate transformation between images using the projective group. M.I.T Media Laboratory Perceptual Computing Section Technical Report No. 278.
[6] George Wolberg. Digital Image Warping. IEEE Computer Society Press, 10662 Los Vaqueros Circle, Los Alamitos, CA, 1990. IEEE Computer Society Press Monograph.
[7] W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T. Vetterling. Numerical Recipes in C: The Art of Scientific Computing. Cambridge University Press, Cambridge, England, second edition, 1992.
[8] Zhengyou Zheng, RachidDeriche, Olivier Faugeras, and Quang-Tuan Luong. A robust technique for matching two uncalibrated images through the recovery of the unknown epipolor geometry. Artificial Intelligence 78, Pages 87-119, 1995.
[9] Zhengyou Zhang. Determining the Epipolar Geometry and its Uncertainty: A Review. International Journal of Computer Vision 27(2), pages 161-195, 1998.
[10] Richard I. Hartley. In Defence of the Eight-Point Algorithm. In IEEE Transactions on Pattern Analysis and Machine Intellugence, Vol. 19, N0. 6, pages 580-593, June 1997.
[11] Miroslav Trajkovic and Mark Hedley. Fast Corner Detection. Image and Vision Computing 16, pages 75-87, 1998.
[12] P. R. Wolf. Elements of photogrammetry. McGraw-Hill, New York, 1983.
[13] Alan Watt, Mark Watt. Advanced Animation and Rendering Techniques. Theory and Practice. Addison-Wesley, 1995.
[14] Richard I. Hartley. Self-Calibration from Multiple Views with a Rotating Camera. In Third European Conference on Computer Vision (ECCV''94), volumn 1, pages 471-478, May 1994.
[15] R. Kumar, P. Anandan, M. Irani, J. Bergen, and K. Hanna. Representation of Scenes from Collections of Images. In IEEE Workshop on Representations of Visual Scenes, pages 10-17, Cambridge, Massachusetts, June 1995.
[16] Shmuel Peleg and Joshua Herman, Panoramic Mosaicing with VideoBrush. DARPA Image Understanding Workshop, May 1997, pages 261-264.