科技通报
科技通報
과기통보
BULLETIN OF SCIENCE AND TECHNOLOGY
2015年
8期
219-221
,共3页
移动终端%嵌入式%人像识别
移動終耑%嵌入式%人像識彆
이동종단%감입식%인상식별
mobile terminal%embedded%portrait identification
针对视觉传感网络的人像识别过程中容易出现移动设备使用的光照、背景易发生改变,使用者姿态、眼睛、表情等发生变化的问题,提出一种基于移动终端嵌入式的视觉传感网络的人像识别技术,寻找每个样本点的若干个近邻点,通过各样本点的近邻点求出该样本点的局部重建权值矩阵,从而求出该样本点的输出值.对视觉传感网络人像灰度值的均值以及方差进行灰度归一化处理,使其相等.根据视觉传感网络人像特征矢量样本获取两个最佳鉴别矢量,将其作为基底矢量转换成子空间,多次输入视觉传感器网络的人像,获取该类模式的人像特征矢量样本集,给出正态模式下的Bayes分类模型,获取单元型最小距离分类器.仿真实验结果表明,所提方法具有很高的人像识别精度.
針對視覺傳感網絡的人像識彆過程中容易齣現移動設備使用的光照、揹景易髮生改變,使用者姿態、眼睛、錶情等髮生變化的問題,提齣一種基于移動終耑嵌入式的視覺傳感網絡的人像識彆技術,尋找每箇樣本點的若榦箇近鄰點,通過各樣本點的近鄰點求齣該樣本點的跼部重建權值矩陣,從而求齣該樣本點的輸齣值.對視覺傳感網絡人像灰度值的均值以及方差進行灰度歸一化處理,使其相等.根據視覺傳感網絡人像特徵矢量樣本穫取兩箇最佳鑒彆矢量,將其作為基底矢量轉換成子空間,多次輸入視覺傳感器網絡的人像,穫取該類模式的人像特徵矢量樣本集,給齣正態模式下的Bayes分類模型,穫取單元型最小距離分類器.倣真實驗結果錶明,所提方法具有很高的人像識彆精度.
침대시각전감망락적인상식별과정중용역출현이동설비사용적광조、배경역발생개변,사용자자태、안정、표정등발생변화적문제,제출일충기우이동종단감입식적시각전감망락적인상식별기술,심조매개양본점적약간개근린점,통과각양본점적근린점구출해양본점적국부중건권치구진,종이구출해양본점적수출치.대시각전감망락인상회도치적균치이급방차진행회도귀일화처리,사기상등.근거시각전감망락인상특정시량양본획취량개최가감별시량,장기작위기저시량전환성자공간,다차수입시각전감기망락적인상,획취해류모식적인상특정시량양본집,급출정태모식하적Bayes분류모형,획취단원형최소거리분류기.방진실험결과표명,소제방법구유흔고적인상식별정도.
As recognition for visual sensor networks prone to mobile devices using light, in the process of background chang-es, user profile, eyes, facial expression change, such as problems, put forward a kind of mobile terminal based on embedded vision sensing network recognition technology of the portrait, looking for a number of each sample point is the neighbor points, through the sample points of neighbor points out local reconstruction weights matrix of the sample points, thus the output value of the sample points. Of visual sensor networks like grey value of the mean and variance of gray scale normal-ization processing, so that it is equal. Root based on the visual sensing network as characteristic vector samples for obtain-ing the best identify two vectors, as a basal vector into subspace, portrait, multiple input visual sensor network access to the class as characteristic vector of pattern sample set, given normal mode of the Bayes classification model, the minimum dis-tance classifier for unit type. The simulation results show that the proposed method has the very high as the identification accuracy.