對稱局部保持的半監(jiān)督維數(shù)約簡算法
發(fā)布時間:2018-11-16 19:02
【摘要】:針對自然界較多圖像具有對稱的特點(diǎn)以及數(shù)據(jù)分布大多呈一定的流形結(jié)構(gòu)情況,提出了一種對稱局部保持的半監(jiān)督維數(shù)約減(SLPSDR)算法.該算法使用矩陣定義維數(shù)約減映射矩陣元素之間的關(guān)系,使圖像中對稱的像素點(diǎn)對應(yīng)的映射矩陣的值之間的差別最小;同時為了利用無標(biāo)簽訓(xùn)練樣本保持?jǐn)?shù)據(jù)的流形結(jié)構(gòu),要求低維空間中每個點(diǎn)的鄰域關(guān)系與高維空間中的鄰域關(guān)系相似.在CMU PIE、Extend Yale B、ORL、AR人臉數(shù)據(jù)庫上的實(shí)驗(yàn)結(jié)果表明,圖像數(shù)據(jù)明顯的對稱特點(diǎn)使得SLPSDR算法優(yōu)于其他對比的維數(shù)約減算法.
[Abstract]:In view of the symmetry of many images in nature and the fact that most of the data distribution is manifold structure, a semi-supervised dimension reduction (SLPSDR) algorithm with symmetric local preservation is proposed. In this algorithm, the matrix is used to define the relationship between the elements of the dimensionality reduction mapping matrix, and the difference between the values of the mapping matrix corresponding to the symmetric pixel points in the image is minimized. At the same time, in order to use unlabeled training samples to maintain the manifold structure of data, it is required that the neighborhood relationship of each point in low-dimensional space is similar to that in high-dimensional space. The experimental results on CMU PIE,Extend Yale ORL AR face database show that the obvious symmetry of the image data makes the SLPSDR algorithm superior to other contrast dimension reduction algorithms.
【作者單位】: 廣東司法警官職業(yè)學(xué)院信息管理系;
【基金】:國家自然科學(xué)基金資助項(xiàng)目(61402118)~~
【分類號】:TP391.41
本文編號:2336374
[Abstract]:In view of the symmetry of many images in nature and the fact that most of the data distribution is manifold structure, a semi-supervised dimension reduction (SLPSDR) algorithm with symmetric local preservation is proposed. In this algorithm, the matrix is used to define the relationship between the elements of the dimensionality reduction mapping matrix, and the difference between the values of the mapping matrix corresponding to the symmetric pixel points in the image is minimized. At the same time, in order to use unlabeled training samples to maintain the manifold structure of data, it is required that the neighborhood relationship of each point in low-dimensional space is similar to that in high-dimensional space. The experimental results on CMU PIE,Extend Yale ORL AR face database show that the obvious symmetry of the image data makes the SLPSDR algorithm superior to other contrast dimension reduction algorithms.
【作者單位】: 廣東司法警官職業(yè)學(xué)院信息管理系;
【基金】:國家自然科學(xué)基金資助項(xiàng)目(61402118)~~
【分類號】:TP391.41
【相似文獻(xiàn)】
相關(guān)期刊論文 前4條
1 鄭誠,歐陽為民,蔡慶生;一種有效的的時間序列維數(shù)約簡方法[J];小型微型計算機(jī)系統(tǒng);2002年11期
2 劉麗娜;于治樓;張化祥;;針對不平衡數(shù)據(jù)集的維數(shù)約簡方法[J];信息技術(shù)與信息化;2011年05期
3 陳威;;幾種維數(shù)約簡方法在人臉識別中的比較研究[J];甘肅聯(lián)合大學(xué)學(xué)報(自然科學(xué)版);2011年04期
4 ;[J];;年期
相關(guān)博士學(xué)位論文 前1條
1 殷飛;基于稀疏學(xué)習(xí)的圖像維數(shù)約簡和目標(biāo)識別方法研究[D];西安電子科技大學(xué);2014年
相關(guān)碩士學(xué)位論文 前1條
1 許新忠;基于維數(shù)約簡的流形可視化研究[D];中國民航大學(xué);2014年
,本文編號:2336374
本文鏈接:http://sikaile.net/kejilunwen/ruanjiangongchenglunwen/2336374.html
最近更新
教材專著