Active Object Movie using Temporal Consistency and Occlusion Recovery
碩士 === 國立東華大學 === 資訊工程學系 === 92 === Active object movie is an object movie with movable objects. The technique of active object movie allows users to change their view-point freely when they are watching a object video sequence. The feasibility of an active object movie is measured by the throug...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | zh-TW |
Published: |
2004
|
Online Access: | http://ndltd.ncl.edu.tw/handle/52672842765499166169 |
id |
ndltd-TW-092NDHU5392054 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-092NDHU53920542016-06-17T04:16:18Z http://ndltd.ncl.edu.tw/handle/52672842765499166169 Active Object Movie using Temporal Consistency and Occlusion Recovery 動態物件電影中時間關聯性的利用與遮蔽區域的還原 Shih-Yu Huang 黃士育 碩士 國立東華大學 資訊工程學系 92 Active object movie is an object movie with movable objects. The technique of active object movie allows users to change their view-point freely when they are watching a object video sequence. The feasibility of an active object movie is measured by the throughput of virtual view-point image synthesizing and the quality of the synthesized image. In video sequences with real-time frame rate, the visual contents are usually similar between consecutive frames. This kind of correlation is referred as temporal consistency. By analyzing the difference between consecutive frames, we can segment the image into four regions : moving, uncorresponding, invalidated ,and heritable. Exploiting temporal consistency between consecutive frames can reduce matching costs, and then virtual view-point images can be synthesized more efficiently. The quality of the synthesize image is depending on the accuracy of the disparity estimation. We estimate dense disparity by stereo matching. Due to occlusion, those pixels that cannot find corresponding points are referred as holes in the disparity map. We analyze the disparity distributions nearby every holds and categorize then into three cases. We propose interpolation strategies for each case to recover the holes in the disparity map. The experiment shows that stereo matching using temporal consistency is twice faster than traditional methods, and occlusion recovery can decrease the SSD error in the virtual view-point image. Mau-Tsuen Yang 楊茂村 2004 學位論文 ; thesis 59 zh-TW |
collection |
NDLTD |
language |
zh-TW |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 國立東華大學 === 資訊工程學系 === 92 ===
Active object movie is an object movie with movable objects. The technique of active object movie allows users to change their view-point freely when they are watching a object video sequence. The feasibility of an active object movie is measured by the throughput of virtual view-point image synthesizing and the quality of the synthesized image.
In video sequences with real-time frame rate, the visual contents are usually similar between consecutive frames. This kind of correlation is referred as temporal consistency. By analyzing the difference between consecutive frames, we can segment the image into four regions : moving, uncorresponding, invalidated ,and heritable. Exploiting temporal consistency between consecutive frames can reduce matching costs, and then virtual view-point images can be synthesized more efficiently.
The quality of the synthesize image is depending on the accuracy of the disparity estimation. We estimate dense disparity by stereo matching. Due to occlusion, those pixels that cannot find corresponding points are referred as holes in the disparity map. We analyze the disparity distributions nearby every holds and categorize then into three cases. We propose interpolation strategies for each case to recover the holes in the disparity map.
The experiment shows that stereo matching using temporal consistency is twice faster than traditional methods, and occlusion recovery can decrease the SSD error in the virtual view-point image.
|
author2 |
Mau-Tsuen Yang |
author_facet |
Mau-Tsuen Yang Shih-Yu Huang 黃士育 |
author |
Shih-Yu Huang 黃士育 |
spellingShingle |
Shih-Yu Huang 黃士育 Active Object Movie using Temporal Consistency and Occlusion Recovery |
author_sort |
Shih-Yu Huang |
title |
Active Object Movie using Temporal Consistency and Occlusion Recovery |
title_short |
Active Object Movie using Temporal Consistency and Occlusion Recovery |
title_full |
Active Object Movie using Temporal Consistency and Occlusion Recovery |
title_fullStr |
Active Object Movie using Temporal Consistency and Occlusion Recovery |
title_full_unstemmed |
Active Object Movie using Temporal Consistency and Occlusion Recovery |
title_sort |
active object movie using temporal consistency and occlusion recovery |
publishDate |
2004 |
url |
http://ndltd.ncl.edu.tw/handle/52672842765499166169 |
work_keys_str_mv |
AT shihyuhuang activeobjectmovieusingtemporalconsistencyandocclusionrecovery AT huángshìyù activeobjectmovieusingtemporalconsistencyandocclusionrecovery AT shihyuhuang dòngtàiwùjiàndiànyǐngzhōngshíjiānguānliánxìngdelìyòngyǔzhēbìqūyùdeháiyuán AT huángshìyù dòngtàiwùjiàndiànyǐngzhōngshíjiānguānliánxìngdelìyòngyǔzhēbìqūyùdeháiyuán |
_version_ |
1718307385967116288 |