华南理工大学学报(自然科学版)
華南理工大學學報(自然科學版)
화남리공대학학보(자연과학판)
Journal of South China University of Technology (Natural Science Edition)
2015年
9期
88-94
,共7页
谭飞刚%刘伟铭%黄玲%翟聪
譚飛剛%劉偉銘%黃玲%翟聰
담비강%류위명%황령%적총
加权欧氏距离%目标再识别%相似性度量%人体再识别%显著性LBP 特征
加權歐氏距離%目標再識彆%相似性度量%人體再識彆%顯著性LBP 特徵
가권구씨거리%목표재식별%상사성도량%인체재식별%현저성LBP 특정
weighted Euclidean distance%object re-identification%similarity measure%person re-identification%significant local binary pattern
针对传统欧氏距离在特征相似性度量中存在区分能力弱的缺陷,提出了基于加权欧氏距离度量的目标再识别算法。首先,针对现有目标再识别算法中目标分割易受衣着和背景颜色干扰的缺陷以及忽略人体头部特征的现象,提出了一种简单的比例分割方法,即根据 VIPeR 和 i-LIDS 数据集上目标各部件的比例统计将目标按比例分割成3部分。然后提取各部件的多种互补特征来增加其对光照变化等因素的鲁棒性。在部件特征描述过程中,文中提出了以显著性因子为权重的显著性局部二值模式(SLBP)特征来增加局部二值模式(LBP)特征对目标显著性的描述。最后综合各部件的相似性度量结果来判断目标是否匹配。在 VIPeR 和 i-LIDS 数据集上的对比实验结果显示,文中算法的目标再识别准确率优于其他算法。
針對傳統歐氏距離在特徵相似性度量中存在區分能力弱的缺陷,提齣瞭基于加權歐氏距離度量的目標再識彆算法。首先,針對現有目標再識彆算法中目標分割易受衣著和揹景顏色榦擾的缺陷以及忽略人體頭部特徵的現象,提齣瞭一種簡單的比例分割方法,即根據 VIPeR 和 i-LIDS 數據集上目標各部件的比例統計將目標按比例分割成3部分。然後提取各部件的多種互補特徵來增加其對光照變化等因素的魯棒性。在部件特徵描述過程中,文中提齣瞭以顯著性因子為權重的顯著性跼部二值模式(SLBP)特徵來增加跼部二值模式(LBP)特徵對目標顯著性的描述。最後綜閤各部件的相似性度量結果來判斷目標是否匹配。在 VIPeR 和 i-LIDS 數據集上的對比實驗結果顯示,文中算法的目標再識彆準確率優于其他算法。
침대전통구씨거리재특정상사성도량중존재구분능력약적결함,제출료기우가권구씨거리도량적목표재식별산법。수선,침대현유목표재식별산법중목표분할역수의착화배경안색간우적결함이급홀략인체두부특정적현상,제출료일충간단적비례분할방법,즉근거 VIPeR 화 i-LIDS 수거집상목표각부건적비례통계장목표안비례분할성3부분。연후제취각부건적다충호보특정래증가기대광조변화등인소적로봉성。재부건특정묘술과정중,문중제출료이현저성인자위권중적현저성국부이치모식(SLBP)특정래증가국부이치모식(LBP)특정대목표현저성적묘술。최후종합각부건적상사성도량결과래판단목표시부필배。재 VIPeR 화 i-LIDS 수거집상적대비실험결과현시,문중산법적목표재식별준학솔우우기타산법。
As the traditional Euclidean distance has a weak distinctive ability in the feature similarity measure,an object re-identification algorithm based on the weighted Euclidean distance metric is proposed.First,aiming at the problems of the existing object re-identification algorithm,which are that the object segmentation is sensitive to clothing and background color and the human head information is ignored,a simple segmentation method is pro-posed,which divides a person into three parts according to the statistics of the proportion of each part in VIPeR and i-LIDS data-sets.Then,various complementary features of each part are extracted to improve the robustness of the proposed algorithm to illumination changes and other factors.A significant local binary pattern (SLBP)with a sig-nificant factor as the weight is proposed to increase the description ability of the local binary pattern (LBP)to the significance of the object in the part feature description process.Finally,the comprehensive result of the similarity measure of each part is used to determine whether the object is matched.The results of comparative experiments on VIPeR and i-LIDS datasets show that the proposed algorithm is superior to other algorithms in terms of accuracy.