Full metadata record
DC FieldValueLanguage
dc.contributor.author詹翊瑄zh_TW
dc.contributor.author蕭旭峰zh_TW
dc.contributor.authorZhan, Yi-Xuanen_US
dc.date.accessioned2018-01-24T07:40:33Z-
dc.date.available2018-01-24T07:40:33Z-
dc.date.issued2017en_US
dc.identifier.urihttp://etd.lib.nctu.edu.tw/cdrfb3/record/nctu/#GT070356535en_US
dc.identifier.urihttp://hdl.handle.net/11536/141367-
dc.description.abstract在論文中,我們提出階層式超像素(superpixel)組成超體素(supervoxel)切割之影片切割方法。在圖像切割方面,我們基於用在超像素切割上的lazy random walk (LRW)演算法,藉由調整超像素種子的中心點再定位,以及修改原本的分割大塊超像素條件等方法,讓超像素的切割結果得到改善。在影片切割方面,我們將超像素切割得到的結果當作所提出之超體素切割方法的輸入,進行所階層式超像素組成超體素演算法,得到最後的超體素結果。 跟原本的LRW超像素切割方法以及其他方法比較,我們提出的方法所得到的超像素結果有較佳的表現。在影片切割方面,我們提出的方法跟其他方法比較,在計算分割體積的正確性和分割錯誤程度性上有不錯的表現。zh_TW
dc.description.abstractIn this thesis, we present a supervoxel approach for video segmentation using a hierarchical architecture of lazy random walk (LRW) algorithm. In image segmentation, we modify the original LRW for superpixel segmentation method and get better results of the superpixels. We improve the superpixel optimization in the original method by modifying the center relocation procedure of the superpixel seeds. A new splitting strategy is further proposed using dynamic threshold approach and better metrics of splitting conditions. In video segmentation, we take the superpixel results from video sequences as spatiotemporal inputs for supervoxel segmentation using the proposed hierarchical LRW supervoxel approach. Comparing with the original LRW superpixel method and another method, our superpixel results have better performance; for video segmentation, our supervoxel results perform well on the 3D segmentation accuracy and 3D undersegmentation error.en_US
dc.language.isoen_USen_US
dc.subject超像素zh_TW
dc.subject超體素zh_TW
dc.subject圖像切割zh_TW
dc.subject影片切割zh_TW
dc.subjectsuperpixelen_US
dc.subjectsupervoxelen_US
dc.subjectimage segmentationen_US
dc.subjectvideo segmentationen_US
dc.subjectlazy random walksen_US
dc.title階層式超像素組成超體素切割方法zh_TW
dc.titleHierarchical Superpixels-to-Supervoxels Segmentation Approachen_US
dc.typeThesisen_US
dc.contributor.department網路工程研究所zh_TW
Appears in Collections:Thesis