Shin color is one of most important human face characteristic , the gaussian model skin color segmentation under ycbcr chrominance space method is choose by the thesis 肤色是人脸重要特征,在通过肤色采样统计和聚类分析后,确立一种在ycbcr空间下的基于高斯模型的肤色分割方法。
Background modeling was based on mixture gaussian model , which had strong adaptive capability for variety of external factors , such as illumination change , shadow , object shelter and etc 采用混合高斯模型为码头背景建模，该模型可克服各种外部因素的影响，如光照变化、阴影、目标遮挡等，具有很强的适应能力。
The improved algorithm of motion detection based on mixture of gaussians is proposed to eliminate the " bad - tail " appearance occurring in background subtraction of slowly moving and large objects 提出改进的基于混合高斯模型的运动检测算法，有效抑制了分割运动缓慢且成像面积比较大的物体时易出现的“拖尾”现象。
The first step is construct the gaussian model for skin color under ycbcr and get the likely region of face , then skin color segmentation is carry out with the method of dynamic threshold optimizing 在ycbcr色彩空间中建立肤色分布的高斯模型,得到肤色概率似然图像,在最佳动态阈值选取算法下完成肤色区域的分割。
The kalman filter is widely applied in the information fusion at the present , which can get the optimal estimate in the linear - gaussian model , but not applied in the nonlinear and non - gaussian model 目前在信息融合领域广泛使用的融合算法是卡尔曼滤波,它在线性高斯模型下能得到最优估计,但在非线性非高斯模型下则无法应用。
The author used an expected sufficient statistics update equations to begin their estimating of the gaussian mixture model , then switch to l - recent window version when the first l samples are processed , and got a good performance 作者用一系列的修正方程式统计数据开始了他们对混合高斯模型的估算，然后将初步得出的数据转换为滤镜视窗版本进行查看，效果还不错。
There are three steps of the concrete process of the algorithm : initial localization , separating image and confirmation . firstly , single gaussian model is used to imitate skin - color distribution , meanwhile , skin probability images which is smoothly disposed using middle wave is calculated to attain the initial localization . secondly , in order to separate the latent face region , it uses multimodal information that includes the region growing 、 the hue and brightness 、 skin probability images etc . at last , it adopts the criterion of morphology analysis and face characteristic draw to confirm the latent region 算法的具体过程主要分为三个阶段:人脸初定位、人脸分割和人脸确认,首先采用高斯模型来模拟肤色分布并构造出肤色概率图进行人脸的初定位,并对肤色概率图采用中值滤波进行平滑处理,接着利用区域生长准则和融合色度、亮度以及肤色概率图等多源信息相结合的方法对图像进行分割,以达到分割出潜在脸区的目的,最后采用人脸整体形状的确认准则和人脸特征提取的确认准则相结合的方法对潜在脸区的最终确认。
In order to make the mgb more easy - operating , ultrasonic testing simula - tion software was designed and developed , which has a good user - oriented interface and can simulate easily the inspection fields of planar transducer , focused transducer and angle beam transducer . it also can help inspectors analyze the inspection procedures and predict the feasibility of actual ultrasonic testing , then choose suitable inspection parameters . the side - drill hole and flat - bottom hole are commonly used in ultrasonic 为了使多元高斯模型使用更加方便,本文设计并开发了基于多元高斯声束理论的超声检测声场模拟软件,该软件具有良好的图形用户界面,操作简单,可方便地实现直探头、聚焦探头和斜探头的检测声场模拟,可帮助检测人员进行检测过程的分析及结构件的可检性预测,并选择合适的检测工艺参数。
In the non - gaussian filtering algorithm , the measurement errors incorporating the multipath effects are modeled as the non - gaussian noise , and the filter are modified according to the score function . this algorithm , designed to track low - elevation targets , avoids the degrading of me performance because of ; . . c non - gaussian noise in the filter 在非高斯算法中，由于多径误差的影响，目标高度方向上的测量误差表现为非高斯噪声，首先建立观测噪声的非高斯模型，然后利用非高斯噪声下的滤波算法跟踪低空目标。
Meanwhile , we present the method of designing multi - class classifier . to solve the problem of some images ’ poor quality , we give the standard to estimate the performance of utterance rejection method : utterance rejection accuracy and recall rate . and we bring forward the utterance rejection method based on voting threshold and analyze the performance of our algorithm 为了比较svm的识别性能,本文也采用了基于混合高斯模型的车牌字符识别算法,给出了基于gmm的分类器的训练和识别的算法,提出了基于概率密度的拒识算法,并就识别准确率和拒识性能与svm进行了比较。
