基于深度数据的空间人脸旋转角度估计

提出一种基于三维人脸深度数据的人脸姿态计算方法。利用人脸的深度数据以及与其一一对应的灰度图像,根据微分几何原理和相应的曲率算法与人脸数据中的灰度特征对人脸面部关键特征点定位,进而计算出人脸姿态在三维空间中的3个姿态角。实验证明该方法能在姿态变化情况下实现对人脸旋转角的准确估计,为进一步的人脸识别和表情分析提供基础。     

基于稀疏模型的人脸姿态估计

针对现有的人脸姿态估计方法易受"自遮挡"影响,采用改进的ASM算法提取人脸特征点,并利用人脸形态的几何统计知识来估计人脸特征点的深度值。以人脸主要特征点建立人脸稀疏模型,在利用相关人脸特征点近似估计人脸姿态后,通过最小二乘法精确估计三维人脸空间姿态。实验结果表明,对于"自遮挡"情况,该方法仍有较好的估计结果,与同类方法比较具有良好的姿态估计精度。     

级联网络和金字塔光流的旋转不变人脸检测

在无约束的开放空间中,由于面部姿态变化、背景环境复杂、运动模糊等,人脸检测仍是一个具有挑战性的任务。本文针对视频流中人脸检测存在的平面内旋转问题,将人脸关键点与金字塔光流相结合,提出了基于级联网络和金字塔光流的旋转不变人脸检测算法。首先利用级联渐进卷积神经网络对视频流中前一帧进行人脸位置和关键点的定位;其次为获取关键点与人脸候选框间光流映射,使用独立的关键点检测网络对当前帧进行再次定位;之后计算前后两帧之间关键点光流位移;最后通过关键点光流位移与人脸候选框的映射关系,对视频中检测到的人脸进行校正,从而完成平面内旋转人脸不变性检测。实验经FDDB公开数据集上测试,证明该方法精确度较高。并且,在Boston面部跟踪数据集上进行动态测试,证明该人脸检测算法能有效解决平面内旋转人脸检测问题。对比其它检测算法,该算法检测速度有较大优势,同时视频中窗口抖动问题得到了很好解决。     

基于深度学习的颜值估计与电商精准营销

近年来,随着机器学习和人工智能领域的不断发展,使得人脸颜值估计的研究得到广泛关注。提出一种基于深度学习的颜值估计框架,利用深度卷积神经网络(Deep Convolutional Neural Network,DCNN)提取人脸图像的特征向量,并采用回归分析计算方法评估人脸颜值,为充分发挥深度卷积神经网络提取特征的能力,提出了优化后的人脸特征提取损失函数。最后,利用该颜值估计算法构建消费者颜值与服装购物偏好相关性模型。结果显示:消费者颜值与服装购物偏好存在一定的相关关系,即颜值越高的消费者越喜欢时尚款式和风格的服装。研究结论为电商企业设计出高度精准营销策略,输出个性化产品和服务提供可能。     

基于特征直线的目标被动定位方法

提出一种利用导引头图像信息进行飞机目标被动定位的方法.该方法以目标可预知的先验几何特征为基础,并基于摄像系统视场较小且目标距离远大于目标尺寸时,成像模型可用弱透视投影近似的事实,推导出借助特征直线估计目标距离的单目视觉算法.与传统的目标定位方法不同,该算法通过特征直线的成像尺寸以及相互夹角反映目标成像姿态的变化,使得目标定位算法不依赖于对目标姿态的估计,提高了不同的空间姿态下目标定位的准确性.实验结果表明,该算法能够在低信噪比情况下准确定位目标,且算法不受目标姿态的影响.     

基于三维模型的多姿态人脸识别

提出了一种利用三维人脸模型匹配二维人脸图像的分层人脸识别方法和基于模糊数学的人脸姿态角度估计算法.对多姿态二维图像进行姿态空间划分,利用主成分分析方法(PCA)形成多姿态特征脸.识别过程首先估计测试图像姿态和模糊姿态角,在估计的姿态空间内采用基于PCA的方法进行第一层识别得到候选个体,然后利用候选个体的三维模型结合模糊姿态角产生虚拟图像,利用相关进行第二层识别.实验结果表明,该方法对姿态的变化有较好的鲁棒性.     

多次投影算法在视频监控人脸检测中的应用

根据视频监控图像在时间上的连续性和空间上的继承性,利用连续三帧视频图像对称差分,找到运动区域,再结合人脸肤色的聚类特征确定出人脸候选区域,然后改进了利用投影的人脸定位算法,将单次投影发展为多次投影,并且结合人脸的几何特征,实现视频监控中复杂背景下的多人脸检测。实验表明,该算法复杂度小,准确率较高,对姿态、表情、背景等变化情况下人脸的检测均具有较好的鲁棒性。     

运用肤色信息和模板匹配的彩色人脸检测

人脸是一个复杂的模式,在图像中自动地对其进行定位和分割是进行人脸识别的第一步。本文提出一种运用肤色信息和模板匹配的人脸检测方法。该方法先进行肤色分割,然后对每一个人脸候选区域进行形状比例的分析,最后进行模板匹配。实验结果表明,该方法对任意背景下,任意姿态及任意数目的人脸检测非常有效。     

基于深度学习的动态场景相机姿态估计方法

针对现有增强现实技术中应用较为广泛的基于标识物的定位注册方法的不足,提出了一种在不断变化运动的复杂动态场景下估计相机连续运动的3维姿态的方法。基于深度神经网络对输入图像序列建立端到端的学习模型,将卷积神经网络(CNN)作为高层特征提取器,同时利用长短期记忆神经网络(LSTM)建立视频连续帧之间的时序关系,完成相机连续运动的3维姿态估计,从而避免了相机快速运动及场景不断运动变换导致图像特征提取效果不好的情况。另一方面,通过迁移学习的方法来预测未知视频序列的相机3维姿态信息,解决了原始数据量不够的问题。在公共数据集上的实验结果表明,相对于PoseNet,基于连续视频序列的输入,其预测精度得到一定的提升。     

基于人脸肤色和特征的实时检测跟踪算法

针对实时视频中的运动人体目标,提出一种基于人脸肤色和特征的快速人脸检测和跟踪方法.首先运用帧差法和形状信息检测出视频中的运动人体目标;然后在YCbVCr色彩空间中,根据肤色的色度聚类特性建立Gaussian模型,分割出肤色区域,去除噪声后,结合人脸的几何特征和器官独有的颜色特征滤除非人脸肤色区域,准确定位人脸在图像中的位置;利用控制策略驱动摄像机,根据人脸信息使人的头肩部位始终处于视频图像的中心,从而实现运动跟踪.为了增强系统对光线变化的适应性,提出了适当的Gaussian模型参数更新策略.实验结果表明,该算法能够适应复杂背景下的人脸检测,具有速度快、准确率高、鲁棒性好的特点,实现了运动人脸的可靠跟踪.     

基于核主元分析的头部姿势估计

现有头部姿势估计方法主要是基于几何分析和基于外现线性变换的方法,计算复杂、通用性不强.提出一种新的利用非线性的核变换算法进行姿势估计的方法,根据流形学习理论,不同姿势的高维人脸图像存在一低维流形结构,提取该流形结构可估计头部姿势.核主元分析是一种非线性降维算法,能够把这种流形结构嵌入到低维空间.利用核主元分析训练姿势估计曲线,然后把新图像投影到姿势曲线上,利用插值方法估计新投影点对应得姿势角度.核主元分析的方法克服了传统线性估计方法的缺点,实验证明该方法估计效果良好,并给出进一步提高估计效果的途径.     

一种高效的实时视觉姿态测量

由光学传感器和计算机视觉理论来确定目标物体的姿态参数具有重要的实用价值和理论意义.本文以提高当前物体姿态求解方法的实时性和准确性为目的,提出一种高效的基于特征点的姿态求解方法,该方法首先采用P3P算法求解目标物体的近似姿态,然后将该近似姿态作为初始值嵌入正交迭代(OI)优化算法求取目标姿态的最优值,该方法可以适用于任意...     

Race Classification Using Deep Learning

Race classification is a long-standing challenge in the field of face image analysis. The investigation of salient facial features is an important task to avoid processing all face parts. Face segmentation strongly benefits several face analysis tasks, including ethnicity and race classification. We propose a race-classification algorithm using a prior face segmentation framework. A deep convolutional neural network (DCNN) was used to construct a face segmentation model. For training the DCNN, we label face images according to seven different classes, that is, nose, skin, hair, eyes, brows, back, and mouth. The DCNN model developed in the first phase was used to create segmentation results. The probabilistic classification method is used, and probability maps (PMs) are created for each semantic class. We investigated five salient facial features from among seven that help in race classification. Features are extracted from the PMs of five classes, and a new model is trained based on the DCNN. We assessed the performance of the proposed race classification method on four standard face datasets, reporting superior results compared with previous studies.     

Interest points localization for brain image using landmark‐annotated atlas

The localization of clinically important points in brain images is crucial for many neurological studies. Conventional manual landmark annotation requires expertise and is often time‐consuming. In this work, we propose an automatic approach for interest point localization in brain image using landmark‐annotated atlas (LAA). The landmark detection procedure is formulated as a problem of finding corresponding points of the atlas. The LAA is constructed from a set of brain images with clinically relevant landmarks annotated. It provides not only the spatial information of the interest points of the brain but also the optimal features for landmark detection through a learning process. Evaluation was performed on 3D magnetic resonance (MR) data using cross‐validation. Obtained results demonstrate that the proposed method achieves the accuracy of ～ 2 mm, which outperforms the traditional methods such as block matching technique and direct image registration. © 2012 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 22, 145–152, 2012     

Cooperative Tracking Using Multiple PTZ Thermal Imagers

A cooperative object tracking framework is proposed which shifts the priority of tracking by pose estimation based on registration between fields of view (FOVs) of different pan?Ctilt?Czoom (PTZ) thermal imagers, avoiding transferring the local features from one imager to another. When an object is selected for tracking, the related PTZ thermal imager tracks it using an improved particle filtering method, and estimates the pose of the imager simultaneously. Once the object enters an overlapping FOV of two imagers, the handoff thermal imager is activated immediately according to the spatial relationship built by pose estimation, and turns to tacking and pose estimation status. Meanwhile, the previous camera returns to waiting status for reactivation. Experiments are conducted to show the proposed framework is applicable to cooperative object tracking for PTZ thermal infrared imagers.     

Statistical phase-shifting step estimation algorithm based on the continuous wavelet transform for high-resolution interferometry metrology

We propose a statistical phase-shifting estimation algorithm for temporal phase-shifting interferometry (PSI) based on the continuous wavelet transform (CWT). The proposed algorithm explores spatial information redundancy in the intraframe interferogram dataset using the phase recovery property on the power ridge of the CWT. Despite the errors introduced by the noise of the interferogram, the statistical part of the algorithm is utilized to give a sound estimation of the phase-shifting step. It also introduces the usage of directional statistics as the statistical model, which was validated, so as to offer a better estimation compared with other statistical models. The algorithm is implemented in computer codes, and the validations of the algorithm were performed on numerical simulated signals and actual phase-shifted moiré interferograms. The major advantage of the proposed algorithm is that it imposes weaker conditions on the presumptions in the temporal PSI, which, under most circumstances, requires uniform and precalibrated phase-shifting steps. Compared with other existing deterministic estimation algorithms, the proposed algorithm estimates the phase-shifting step statistically. The proposed algorithm allows the temporal PSI to operate under dynamic loading conditions and arbitrary phase steps and also without precalibration of the phase shifter. The proposed method can serve as a benchmark method for comparing the accuracy of the different phase-step estimation methods.     

A novel algorithm for pose estimation based on generalized orthogonal iteration with uncertainty-weighted measuring error of feature points

In this paper, we investigate the effect of anisotropic and correlated non-identical distributions of feature points’ gray level on pose estimation in stereo vision system. The generalized orthogonal iteration (GOI) algorithm for pose estimation with uncertainty-weighted measuring error of feature target is proposed. In this method, the inverse covariance matrix is applied to describe the uncertainty of feature points, and weighted contribution of uncertainty to the error objective function is analysed. By transforming the uncertainty into the covariance-weighted data space, a novel objective function based on spatial collinear error is constructed. The orthogonal iterative algorithm is extended to stereo vision system for pose estimation and the GOI algorithm is deduced, by which the optimized solution to a novel objective function is given. Finally, simulation and actual experimental results show high accuracy and strong robustness of the proposed approach, and should therefore, have potential for a variety of engineering applications.     

Face verification through tracking facial features.

We propose an algorithm for face verification through tracking facial features by using sequential importance sampling. Specifically, we first formulate tracking as a Bayesian inference problem and propose to use Markov chain Monte Carlo techniques for obtaining an empirical solution. A reparameterization is introduced under parametric motion assumption, which facilitates the empirical estimation and also allows verification to be addressed along with tracking. The facial features to be tracked are defined on a grid with Gabor attributes (jets). The motion of facial feature points is modeled as a global two-dimensional (2-D) affine transformation (accounting for head motion) plus a local deformation (accounting for residual motion that is due to inaccuracies in 2-D affine modeling and other factors such as facial expression). Motion of both types is processed simultaneously by the tracker: The global motion is estimated by importance sampling, and the residual motion is handled by incorporating local deformation into the measurement likelihood in computing the weight of a sample. Experiments with a real database of face image sequences are presented.