Novel View Synthesis from Multiple Cameras by Using Dense Correspondence Technique
碩士 === 華梵大學 === 資訊管理學系碩士班 === 96 === The purpose of the thesis is to synthesize a novel view based on multiple images with different view angles. There are two main steps to accomplish the purpose: several images are used for finding dense stereo correspondences, and then the novel views are synthes...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | zh-TW |
Published: |
2008
|
Online Access: | http://ndltd.ncl.edu.tw/handle/37133211767382101518 |
id |
ndltd-TW-096HCHT0396026 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-096HCHT03960262015-10-13T13:47:37Z http://ndltd.ncl.edu.tw/handle/37133211767382101518 Novel View Synthesis from Multiple Cameras by Using Dense Correspondence Technique 從多視角影像產生密集對應點以合成新視點影像 Yi-Huan Liu 劉宜桓 碩士 華梵大學 資訊管理學系碩士班 96 The purpose of the thesis is to synthesize a novel view based on multiple images with different view angles. There are two main steps to accomplish the purpose: several images are used for finding dense stereo correspondences, and then the novel views are synthesized using these correspondences. Refer to stereo correspondences. In our experimental setup, eight cameras which have been well calibrated are used. In the beginning, SIFT (Scale Invariant Feature Transform) is used to obtain good stereo corresponding point pairs which are regarded as seeds. ZNCC (Zero-mean Normalized Cross Correlation) is to filter the bad seeds obtained from using SIFT. The 3D coordinates are then computed using these filtered stereo correspondences in different combinations. In order to improve the correctness of depths generated from stereo correspondences, the 3D points transformed from the world coordinate system into the corresponding camera coordinate systems are projected into other images. Two measurements, such as ZNCC and color histogram, are used to verify the correctness of the corresponding point pairs. Once some sparse 3D points obtained from some corresponding point pairs have been verified, these sparse 3D points are then projected into several images for further processing: a region in an image pair for generating denser stereo correspondences will be selected by using ZNCC. Refer to novel view synthesis. For synthesizing an image from a novel view, the CCV(Color Consistence Verification) will be used to determine the best color value from multiple images. And then, the triangular meshes are generated using Delaunay Triangulation. The interpolation methods are used to synthesize image regions for the regions having no stereo correspondences. Cheng-Yuan Tang 唐政元 2008 學位論文 ; thesis 51 zh-TW |
collection |
NDLTD |
language |
zh-TW |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 華梵大學 === 資訊管理學系碩士班 === 96 === The purpose of the thesis is to synthesize a novel view based on multiple images with different view angles. There are two main steps to accomplish the purpose: several images are used for finding dense stereo correspondences, and then the novel views are synthesized using these correspondences.
Refer to stereo correspondences. In our experimental setup, eight cameras which have been well calibrated are used. In the beginning, SIFT (Scale Invariant Feature Transform) is used to obtain good stereo corresponding point pairs which are regarded as seeds. ZNCC (Zero-mean Normalized Cross Correlation) is to filter the bad seeds obtained from using SIFT. The 3D coordinates are then computed using these filtered stereo correspondences in different combinations. In order to improve the correctness of depths generated from stereo correspondences, the 3D points transformed from the world coordinate system into the corresponding camera coordinate systems are projected into other images. Two measurements, such as ZNCC and color histogram, are used to verify the correctness of the corresponding point pairs. Once some sparse 3D points obtained from some corresponding point pairs have been verified, these sparse 3D points are then projected into several images for further processing: a region in an image pair for generating denser stereo correspondences will be selected by using ZNCC. Refer to novel view synthesis. For synthesizing an image from a novel view, the CCV(Color Consistence Verification) will be used to determine the best color value from multiple images. And then, the triangular meshes are generated using Delaunay Triangulation. The interpolation methods are used to synthesize image regions for the regions having no stereo correspondences.
|
author2 |
Cheng-Yuan Tang |
author_facet |
Cheng-Yuan Tang Yi-Huan Liu 劉宜桓 |
author |
Yi-Huan Liu 劉宜桓 |
spellingShingle |
Yi-Huan Liu 劉宜桓 Novel View Synthesis from Multiple Cameras by Using Dense Correspondence Technique |
author_sort |
Yi-Huan Liu |
title |
Novel View Synthesis from Multiple Cameras by Using Dense Correspondence Technique |
title_short |
Novel View Synthesis from Multiple Cameras by Using Dense Correspondence Technique |
title_full |
Novel View Synthesis from Multiple Cameras by Using Dense Correspondence Technique |
title_fullStr |
Novel View Synthesis from Multiple Cameras by Using Dense Correspondence Technique |
title_full_unstemmed |
Novel View Synthesis from Multiple Cameras by Using Dense Correspondence Technique |
title_sort |
novel view synthesis from multiple cameras by using dense correspondence technique |
publishDate |
2008 |
url |
http://ndltd.ncl.edu.tw/handle/37133211767382101518 |
work_keys_str_mv |
AT yihuanliu novelviewsynthesisfrommultiplecamerasbyusingdensecorrespondencetechnique AT liúyíhuán novelviewsynthesisfrommultiplecamerasbyusingdensecorrespondencetechnique AT yihuanliu cóngduōshìjiǎoyǐngxiàngchǎnshēngmìjíduìyīngdiǎnyǐhéchéngxīnshìdiǎnyǐngxiàng AT liúyíhuán cóngduōshìjiǎoyǐngxiàngchǎnshēngmìjíduìyīngdiǎnyǐhéchéngxīnshìdiǎnyǐngxiàng |
_version_ |
1717741586917359616 |