Neural Network-Based Face Detection

合集下载

基于深度学习的人脸识别技术

基于深度学习的人脸识别技术一、背景介绍人脸识别技术是一种现代化的信息技术，它在安防、智能家居、金融等方面得到了广泛应用。

人脸识别技术的发展历程可以追溯到上个世纪六十年代。

随着计算机的发展和人工智能技术的进步，人脸识别技术也在不断发展。

而基于深度学习的人脸识别技术是当前最先进的人脸识别技术，具有更高的准确性和鲁棒性。

二、基本原理基于深度学习的人脸识别技术的核心是卷积神经网络（Convolutional Neural Network，CNN）。

在人脸识别中，CNN主要实现了两个步骤：人脸检测和人脸识别。

1、人脸检测人脸检测是指在图像或视频流中，通过计算机算法和技术，自动或半自动地找出图像中包含的人脸并进行定位的过程。

在基于深度学习的人脸检测中，主要使用了区域卷积神经网络（Region-based Convolutional Neural Network，R-CNN）和快速区域卷积神经网络（Fast R-CNN）等方法。

2、人脸识别人脸识别是将图像中的人脸进行比对和匹配，从而确定这张人脸的身份的过程。

在基于深度学习的人脸识别中，主要使用了卷积神经网络和循环神经网络（Recurrent Neural Network，RNN）等方法。

三、应用场景基于深度学习的人脸识别技术已经广泛应用于安防、金融、智能家居等领域。

1、安防领域在安防领域中，基于深度学习的人脸识别技术可以实现人员进出监控、黑名单管理、犯罪现场侦查等功能，具有高效、准确、实时、智能的特点。

2、金融领域在金融领域中，基于深度学习的人脸识别技术可以实现账户认证、开户、支付等功能，具有高安全性、高便捷性的特点。

3、智能家居领域在智能家居领域中，基于深度学习的人脸识别技术可以实现人脸门禁、智能家电控制等功能，具有高度个性化、智能化和便捷性的特点。

四、发展前景基于深度学习的人脸识别技术在未来的发展中具有广阔的前景。

随着大数据和人工智能技术的不断发展，基于深度学习的人脸识别技术可以更好地满足实际场景的需求，并不断提高其准确性和鲁棒性。

图像处理和计算机视觉中的经典论文

前言：最近由于工作的关系，接触到了很多篇以前都没有听说过的经典文章，在感叹这些文章伟大的同时，也顿感自己视野的狭小。

想在网上找找计算机视觉界的经典文章汇总，一直没有找到。

失望之余，我决定自己总结一篇，希望对 CV领域的童鞋们有所帮助。

由于自
己的视野比较狭窄，肯定也有很多疏漏，权当抛砖引玉了
1990年之前
1990年
1991年
1992年
1993年
1994年
1995年
1996年
1997年
1998年
1998年是图像处理和计算机视觉经典文章井喷的一年。

大概从这一年开始，开始有了新的趋势。

由于竞争的加剧，一些好的算法都先发在会议上了，先占个坑，等过一两年之后再扩展到会议上。

1999年
2000年
世纪之交，各种综述都出来了
2001年
2002年
2003年
2004年
2005年
2006年
2007年
2008年
2009年
2010年
2011年
2012年。

人脸识别经典—FaceNet

厦门市美亚柏科信息股份有限公司1 / 8 人脸识别经典—FaceNet编者按近年来，随着深度学习的火热发展，人脸识别技术不断创出新高度，更多方法的发掘及大牛们项目的持续开源，让人脸识别、深度学习甚至是人工智能显得不再神秘，LFW 识别率99%已经随处可见，今天美亚柏科技术专家就人脸识别的一些经典算法和资源进行分享，感受大神们的“最强大脑”！在实际应用中，人脸验证（判断是否是同一人）、人脸识别（这个人是谁）和人脸聚类（寻找类似的人）在自然场景应用仍面临一些困难。

为了降低背景和环境等因素带来的干扰，人脸识别一般先经过人脸检测（Face Detection ）、人脸对齐（Face Alignment ）等预处理，然后将人脸图像映射到欧几里得等空间，空间距离的长度代表了人脸图像的相似性。

只要该映射空间生成，人脸识别，验证和聚类等任务就显得不那么复杂。

由于先前的网络和算法在各种测评榜单上不断被”挤下去”，似乎激发不起大伙的兴趣，那么本期我们就一起看看Google 大神的匠心之作——FaceNet。

FaceNet其实就是一个前言所诉的通用人脸识别系统：采用深度卷积神经网络（CNN）学习将图像映射到欧式空间。

空间距离直接和图片相似度相关：同一个人的不同图像在空间距离很小，不同人的图像在空间中有较大的距离，可以用于人脸验证、识别和聚类。

在800万人，2亿多张样本集训练后，FaceNet在LFW数据集上测试的准确率达到了99.63%，在YouTube Faces DB数据集上，准确率为95.12%。

算法作为经典的深度学习、人脸识别案例，依然采用主流的深度神经网络来提取特征，并采用triplet_loss来衡量训练过程中样本之间的距离误差。

在训练前或者在线学习中不断给神经网络制造“困难”，即一直在寻找与样本最不像的“自己”，同时寻找与自己最像的“他人”。

通过随机梯度下降法，不断缩短自身所有样本的差距，同时尽可能拉大与其他人的差距，最终达到一个最优。

人脸识别介绍_IntroFaceDetectRecognition

Knowledge-based Methods: Summary
Pros:
Easy to come up with simple rules Based on the coded rules, facial features in an input image are extracted first, and face candidates are identified Work well for face localization in uncluttered background
Template-Based Methods: Summary
Pros:
Simple
Cons:
Templates needs to be initialized near the face images Difficult to enumerate templates for different poses (similar to knowledgebased methods)
Knowledge-Based Methods
Top Top-down approach: Represent a face using a set of human-coded rules Example:
The center part of face has uniform intensity values The difference between the average intensity values of the center part and the upper part is significant A face often appears with two eyes that are symmetric to each other, a nose and a mouth

人脸图像中眼镜检测与边框去除方法

人脸图像中眼镜检测与边框去除方法陈文青;王佰玲【摘要】眼镜边框是影响精确提取人脸图像特征的因素之一，为此提出了一种眼镜检测和边框去除的方法。

该方法由眼镜检测、眼镜边框定位和被遮挡图像修复三部分构成。

提取眼睛估计区域的边缘特征并基于神经网络的方法检测眼镜；利用二值化和数学形态学的方法定位眼镜边框；通过插值的方法修复被眼镜边框遮挡的图像。

实验结果表明，该方法与传统基于PCA的方法相比，眼镜去除后的人脸图像更加自然。

同时，实验结果也表明该方法有助于人脸识别性能的提升。

%The eyeglasses frame is one important factor affecting the accuracy of face features extraction, so a method of detecting and removing the eyeglasses frame is proposed. The method is composed of glasses detection, frame location and restoral of the occluded images. First, the edge features of eyes region are extracted and the glasses are detected based on neural network. Then, the eyeglasses frame is located based on binarization and morphology. Finally, the image is restored by interpolation method. Experimental results show that compared with the method based on PCA(Principle Component Analysis), the removal eyeglasses face images processed by the proposed method are more natural and can achieve better face recognition performance.【期刊名称】《计算机工程与应用》【年(卷),期】2016(052)015【总页数】6页(P178-182,232)【关键词】眼镜检测;眼镜边框去除;人脸识别;肤色模型【作者】陈文青;王佰玲【作者单位】绍兴职业技术学院信息工程学院，浙江绍兴 312000;哈尔滨工业大学计算机科学与技术学院，哈尔滨 150090【正文语种】中文【中图分类】TP391.41CHEN Wenqing，WANG Bailing.Computer Engineering and Applications，2016，52（15）：178-182.人脸识别是一项重要的生物特征识别技术，在金融、公共安全、人机交互等领域均有广泛的应用。

Neural Network-Based Face Detection

Neural Network-Based Face DetectionHenry A.Rowley har@ /˜har Shumeet Baluja baluja@ /˜baluja School of Computer Science,Carnegie Mellon University5000Forbes Avenue,Pittsburgh,PA 15213,USATakeo Kanade tk@ /˜tkAbstractWe present a neural network-based face detection system.A retinally connected neural network ex-amines small windows of an image,and decides whether each window contains a face.The sys-tem arbitrates between multiple networks to im-prove performance over a single network.We use a bootstrap algorithm for training the networks,which adds false detections into the training set as train-ing progresses.This eliminates the difﬁcult task of manually selecting non-face training examples,which must be chosen to span the entire space of non-face parisons with other state-of-the-art face detection systems are presented;our system has better performance in terms of detection and false-positive rates.1IntroductionIn this paper,we present a neural network-based al-gorithm to detect frontal views of faces in gray-scale images 1.The algorithms and training methods arem Preprocessing Neural networkExtracted window Input image pyramidCorrect lighting Histogram equalization Receptive fieldsFigure 1:The basic algorithm used for face detection.from individual ﬁlters and eliminates overlapping detections.2.1Stage One:A Neural Network-Based FilterThe ﬁrst component of our system is a ﬁlter that receives as input a 20x20pixel region of the image,and generates an output ranging from 1to -1,signi-fying the presence or absence of a face,respectively.To detect faces anywhere in the input,the ﬁlter is applied at every location in the image.To detect faces larger than the window size,the input image is repeatedly reduced in size (by subsampling),and the ﬁlter is applied at each size.The ﬁlter itself must have some invariance to position and scale.The amount of invariance built into the ﬁlter determines the number of scales and positions at which the ﬁlter must be applied.For the work presented here,we apply the ﬁlter at every pixel position in the image,and scale the image down by a factor of 1.2for each step in the pyramid.The ﬁltering algorithm is shown in Figure 1.First,a preprocessing step,adapted from [Sung and Poggio,1994],is applied to a window of the image.The win-dow is then passed through a neural network,which decides whether the window contains a face.The preprocessing ﬁrst attempts to equalize the intensity values across the window.We ﬁt a function which varies linearly across the window to the intensity values in an oval region inside the window.Pixels outside the oval may represent the background,so those intensity values are ignored in computing thelighting variation across the face.The linear func-tion will approximate the overall brightness of each part of the window,and can be subtracted from the window to compensate for a variety of lighting con-ditions.Then histogram equalization is performed,which non-linearly maps the intensity values to ex-pand the range of intensities in the window.The his-togram is computed for pixels inside an oval region in the window.This compensates for differences in camera input gains,as well as improving contrast in some cases.The preprocessed window is then passed through a neural network.The network has retinal connec-tions to its input layer;the receptive ﬁelds of hidden units are shown in Figure 1.There are three types of hidden units:4which look at 10x10pixel sub-regions,16which look at 5x5pixel subregions,and 6which look at overlapping 20x5pixel horizontal stripes of pixels.Each of these types was chosen to allow the hidden units to represent features that might be important for face detection.In particular,the horizontal stripes allow the hidden units to de-tect such features as mouths or pairs of eyes,while the hidden units with square receptive ﬁelds might detect features such as individual eyes,the nose,or corners of the mouth.Although the ﬁgure shows a single hidden unit for each subregion of the input,these units can be replicated.For the experiments which are described later,we use networks with two and three sets of these hidden units.Similar input connection patterns are commonly used in speech and character recognition tasks [Waibel et al.,1989,Le Cun et al.,1989].The network has a single,real-valued output,which indicates whether or not the window contains a face.Examples of output from a single network are shown in Figure 2.In the ﬁgure,each box represents the position and size of a window to which the neural network gave a positive response.The network has some invariance to position and scale,which results in multiple boxes around some faces.Note also that there are some false detections;they will be eliminated by methods presented in Section2.2.A BFigure 2:Images with all the above threshold de-tections indicated by boxes.To train the neural network used in stage one to serve as an accurate ﬁlter,a large number of face and non-face images are needed.Nearly 1050face exam-ples were gathered from face databases at CMU and Harvard 2.The images contained faces of various sizes,orientations,positions,and intensities.The eyes and the center of the upper lip of each face were located manually,and these points were used to normalize each face to the same scale,orientation,and position,as follows:1.The image is rotated so that both eyes appear on a horizontal line.2.The image is scaled so that the distance from the point between the eyes to the upper lip is 12pixels.3.A 20x20pixel region,centered 1pixel above the point between the eyes and the upper lip,isextracted.Figure 3:Example face images,randomly mir-rored,rotated,translated,and scaled by small amounts.Practically any image can serve as a non-face exam-ple because the space of non-face images is much larger than the space of face images.However,col-lecting a “representative”set of non-faces is difﬁ-cult.Instead of collecting the images before training is started,the images are collected during training,in the following manner,adapted from [Sung and Poggio,1994]:1.Create an initial set of non-face images by gen-erating 1000images with random pixel inten-sities.Apply the preprocessing steps to each of these images.2.Train a neural network to produce an output of 1for the face examples,and -1for the non-face examples.The training algorithm is standard error backpropogation.On the ﬁrst iteration of this loop,the network’s weights are initially random.After the ﬁrst iteration,we use theweights computed by training in the previous iteration as the starting point for training.3.Run the system on an image of scenery whichcontains no faces.Collect subimages in which the network incorrectly identiﬁes a face(an output activation0).4.Select up to250of these subimages at random,apply the preprocessing steps,and add them into the training set as negative examples.Go to step2.Some examples of non-faces that are collected dur-ing training are shown in Figure4.We used120 images of scenery for collecting negative examples in this bootstrap manner.A typical training run selects approximately8000non-face images from the146,212,178subimages that are available at all locations and scales in the training scenery images.2.2Stage Two:Merging Overlapping De-tections and ArbitrationThe examples in Figure2showed that the raw output from a single network will contain a number of false detections.In this section,we present two strategies to improve the reliability of the detector:merging overlapping detections from a single network and arbitrating among multiple networks.2.2.1Merging Overlapping Detections Note that in Figure2,most faces are detected at multiple nearby positions or scales,while false de-tections often occur with less consistency.This ob-servation leads to a heuristic which can eliminate many false detections.For each location and scale at which a face is detected,the number of detections within a speciﬁed neighborhood of that location can be counted.If the number is above a threshold,then that location is classiﬁed as a face.The centroid of the nearby detections deﬁnes the location of the detection result,thereby collapsing multiple detec-tions.In the experiments section,this heuristic will be referred to as“thresholding”.If a particular location is correctly identiﬁed as a face,then all other detection locations which over-lap it are likely to be errors,and can therefore be eliminated.Based on the above heuristic regarding nearby detections,we preserve the location with the higher number of detections within a small neigh-borhood,and eliminate locations with fewer ter,in the discussion of the experiments, this heuristic is called“overlap elimination”.There are relatively few cases in which this heuristic fails; however,one such case is illustrated in the left two faces in Figure2B,in which one face partially oc-cludes another.The implementation of these two heuristics is as fol-lows.Each detection by the network at a particular location and scale is marked in an image pyramid, labelled the“output”pyramid.Then,each location in the pyramid is replaced by the number of detec-tions in a speciﬁed neighborhood of that location. This has the effect of“spreading out”the detec-tions.The neighborhood extends an equal number of pixels in the dimensions of scale and position.A threshold is applied to these values,and the centroids (in both position and scale)of all above threshold regions are computed.All detections contributing to the centroids are collapsed down to single points. Each centroid is then examined in order,starting from the ones which had the highest number of de-tections within the speciﬁed neighborhood.If any other centroid locations represent a face overlapping with the current centroid,they are removed from the output pyramid.All remaining centroid locations constitute theﬁnal detection result.2.2.2Arbitration among Multiple Net-worksTo further reduce the number of false positives,we can apply multiple networks,and arbitrate between the outputs to produce theﬁnal decision.Each net-work is trained in the manner described above,but with different random initial weights,random ini-tial non-face images,and random permutations of the order of presentation of the scenery images.As will be seen in the next section,the detection and false positive rates of the individual networks will be quite close.However,because of different training conditions and because of self-selection of negative training examples,the networks will have different biases and will make different errors.Each detection by a network at a particular position and scale is recorded in an image pyramid.OneFigure 4:During training,the partially-trained system is applied to images of scenery which do not contain faces (like the one on the left).Any regions in the image detected as faces (which are expanded and shown on the right)are errors,which can be added into the set of negative training examples.way to combine two such pyramids is by ANDing them.This strategy signals a detection only if both networks detect a face at precisely the same scale and position.Due to the biases of the individual networks,they will rarely agree on a false detection of a face.This allows ANDing to eliminate most false detections.Unfortunately,this heuristic can decrease the detection rate because a face detected by only one network will be thrown out.However,we will show later that individual networks can all detect roughly the same set of faces,so that the number of faces lost due to ANDing is small.Similar heuristics,such as ORing the outputs of two networks,or voting among three networks,were also tried.Each of these arbitration methods can be applied before or after the “thresholding”and “over-lap elimination”heuristics.If applied afterwards,we combine the centroid locations rather than actual detection locations,and require them to be within some neighborhood of one another rather than pre-cisely aligned.Arbitration strategies such as ANDing,ORing,or voting seem intuitively reasonable,but perhaps there are some less obvious heuristics that could perform better.In [Rowley et al.,1995],we tested this hy-pothesis by using a separate neural network to ar-bitrate among multiple detection networks.It was found that the neural network-based arbitration pro-duces results comparable to those produced by the heuristics presented earlier.3Experimental ResultsA large number of experiments were performed to evaluate the system.We ﬁrst show an analysis of which features the neural network is using to detect faces,then present the error rates of the system over three large test sets.3.1Sensitivity AnalysisIn order to determine which part of the input im-age the network uses to decide whether the input is a face,we performed a sensitivity analysis using the method of [Baluja and Pomerleau,1995].We collected a positive test set based on the training database of face images,but with different random-ized scales,translations,and rotations than were used for training.The negative test set was built from a set of negative examples collected during the training of an earlier version of the system.Each of the 20x20pixel input images was divided into 1002x2pixel subimages.For each subimage in turn,we went through the test set,replacing that subimage with random noise,and tested the neural network.The resulting sum of squared errors made by the network is an indication of how important that por-tion of the image is for the detection task.Plots of the error rates for two networks we developed are shown in Figure work 1uses two sets of the hidden units illustrated in Figure 1,while Network 2uses three sets.The networks rely most heavily on the eyes,then on the nose,and then on the mouth (Figure 5).Anec-dotally,we have seen this behavior on several realNetwork 2Face at Same ScaleFigure5:Error rates(vertical axis)on a small test resulting from adding noise to various portions of the input image(horizontal plane),for two work1has two copies of the hidden units shown in Figure1(a total of58hidden units and2905connections),while Network2has three copies(a total of78 hidden units and4357connections).test images.Even in cases in which only one eyeis visible,detection of a face is possible,thoughless reliable,than when the entire face is visible.The system is less sensitive to the occlusion of otherfeatures such as the nose or mouth.3.2TestingThe system was tested on three large sets of images,which are completely distinct from the training sets.Test Set A was collected at CMU,and consists of42scanned photographs,newspaper pictures,im-ages collected from the World Wide Web,and digi-tized television pictures.These images contain169frontal views of faces,and require the networks toexamine22,053,12420x20pixel windows.TestSet B consists of23images containing155faces(9,678,084windows);it was used in[Sung and Pog-gio,1994]to measure the accuracy of their system.Test Set C is similar to Test Set A,but contains someimages with more complex backgrounds and with-out any faces,to more accurately measure the falsedetection rate.It contains65images,183faces,and51,368,003windows.3A feature our face detection system has in commonwith many systems is that the outputs are not binary.The neural networkﬁlters produce real values be-tween1and-1,indicating whether or not the input0.750.80.850.90.9511e-071e-061e-050.00010.0010.010.11F r a c t i o n o f F a c e s D e t e c t e dFalse Detections per Windows ExaminedROC Curve for Test Sets A, B, and Czero zeroNetwork 1Network 2Figure 6:The detection rate plotted against false positives as the detection threshold is varied from -1to 1,for two networks.The performance was measured over all images from Test Sets A,B,and work 1uses two sets of the hidden units illus-trated in Figure 1,while Network 2uses three sets.The points labelled “zero”are the zero threshold points which are used for all other works were based on Networks 1and 2,while voting was based on Networks 1,2,and 3.The table shows the percentage of faces correctly de-tected,and the number of false detections over the combination of Test Sets A,B,and C.[Rowley et al.,1995]gives a breakdown of the performance of each of these system for each of the three test sets,as well as the performance of systems using neural networks to arbitration among multiple detection networks.As discussed earlier,the “thresholding”heuristic for merging detections requires two parameters,which specify the size of the neighborhood used in search-ing for nearby detections,and the threshold on the number of detections that must be found in that neighborhood.In Table 1,these two parameters are shown in parentheses after the word “threshold”.Similarly,the ANDing,ORing,and voting arbitra-tion methods have a parameter specifying how close two detections (or detection centroids)must be in order to be counted as identical.Systems 1through 4show the raw performance of the networks.Systems 5through 8use the same networks,but include the thresholding and overlap elimination steps which decrease the number of false detections signiﬁcantly,at the expense of a small de-crease in the detection rate.The remaining systemsall use arbitration among multiple ing arbitration further reduces the false positive rate,and in some cases increases the detection rate slightly.Note that for systems using arbitration,the ratio of false detections to windows examined is extremely low,ranging from 1false detection per 229,556win-dows to down to 1in 10,387,401,depending on the type of arbitration used.Systems 10,11,and 12show that the detector can be tuned to make it more or less conservative.System 10,which uses ANDing,gives an extremely small number of false positives,and has a detection rate of about 78.9%.On the other hand,System 12,which is based on ORing,has a higher detection rate of 90.5%but also has a larger number of false detections.System 11provides a compromise between the two.The differences in performance of these systems can be understood by considering the arbitration strategy.When using ANDing,a false detection made by only one network is suppressed,leading to a lower false positive rate.On the other hand,when ORing is used,faces detected correctly by only one network will be preserved,improving the detection rate.Sys-tem 13,which uses voting among three networks,yields about the same detection rate and lower false positive rate than System 12,which uses ORing of two networks.Based on the results shown in Table 1,we con-cluded that System 11makes an acceptable tradeoff between the number of false detections and the de-tection rate.System 11detects on average 85.4%of the faces,with an average of one false detection per 1,319,03520x20pixel windows examined.Figure 7shows examples output images from System 11.4Comparison to Other Systems[Sung and Poggio,1994]reports a face detection system based on clustering techniques.Their sys-tem,like ours,passes a small window over all por-tions of the image,and determines whether a face exists in each window.Their system uses a su-pervised clustering method with six “face”and six “non-face”clusters.Two distance metrics measure the distance of an input image to the prototype clus-ters.The ﬁrst metric measures the “partial”distance between the test pattern and the cluster’s 75most signiﬁcant eigenvectors.The second distance met-ric is the Euclidean distance between the test patternTable1:Combined Detection and Error Rates for Test Sets A,B,and CFalse False detect Type detects rate00in830992113792.7%1)Network1(2copies of hidden units(52total),2905connections)15461in537513)Network3(2copies of hidden units(524491.3%total),2905connections)25081in331344690.9%5)Network1threshold(2,1)overlapelimination7191in1155765389.5%7)Network3threshold(2,1)overlapelimination10521in789929)Networks1and2AND(0)6687.0%81in103874017485.4%11)Networks1and2threshold(2,2)overlap elimination AND(2)3621in22955613)Networks1,2,3voting(0)overlap5389.5%eliminationB: 2/2/0C: 1/1/0J: 8/7/1A: 57/57/3I: 7/5/0H: 3/3/0K: 14/14/0L: 1/1/0G: 2/1/0M: 1/1/0F: 11/11/0E: 15/15/0Figure7:Output obtained from System11in Table1.For each image,three numbers are shown:the number of faces in the image,the number of faces detected correctly,and the number of false detections. Some notes on speciﬁc images:False detections are present in A and J.Faces are missed in G(babies with ﬁngers in their mouths are not well represented in the training set),I(one because of the lighting,causing one side of the face to contain no information,and one because of the bright band over the eyes),and J (removed because a false detect overlapped it).Although the system was trained only on real faces,hand drawn faces are detected in D.Images A,I,and K were obtained from the World Wide Web,B was scanned from a photograph,C is a digitized television image,D,E,F,H,and J were provided by Sung and Poggio at MIT,G and L were scanned from newspapers,and M was scanned from a printed photograph.and its projection in the75dimensional subspace. These distance measures have close ties with Prin-cipal Components Analysis(PCA),as described in [Sung and Poggio,1994].The last step in their sys-tem is to use either a perceptron or a neural network with a hidden layer,trained to classify points using the two distances to each of the clusters(a total of 24inputs).Their system is trained with4000posi-tive examples and nearly47500negative examples collected in the“bootstrap”manner.In compari-son,our system uses approximately16000positive examples and9000negative examples.Table2shows the accuracy of their system on Test Set B,along with the results of our system using the heuristics employed by Systems10,11,and12 in Table1.In[Sung and Poggio,1994],149faces were labelled in the test set,while we labelled155. Some of these faces are difﬁcult for either system to detect.Based on the assumption that[Sung and Poggio,1994]were unable to detect any of the six additional faces we labelled,the number of missed faces is six more than the values listed in their paper. It should be noted that because of implementation details,[Sung and Poggio,1994]process a slightly smaller number of windows over the entire test set; this is taken into account when computing the false detection rates.Table2shows that for equal num-bers of false detections,we can achieve higher de-tection rates.The main computational cost in[Sung and Poggio, 1994]is in computing the two distance measures from each new window to12clusters.We estimate that this computation requiresﬁfty times as many ﬂoating point operations as are needed to classify a window in our system,in which the main costs are in preprocessing and applying neural networks to the window.Although there is insufﬁcient space to present them here,[Rowley et al.,1995]describes techniques for speeding up our system,based on the work of [Umezaki,1995]on license plate detection.These techniques are related,at a high level,to those pre-sented in[Vaillant et al.,1994].In that work,two networks were used.Theﬁrst network has a single output,and like our system it is trained to produce a maximal positive value for centered faces,and a maximal negative value for non-faces.Unlike our system,for faces that are not perfectly centered,the network is trained to produce an intermediate value related to how far off-center the face is.This net-work scans over the image to produce candidate face locations.It runs quickly because of the network architecture:using retinal connections and shared weights,much of the computation required for one application of the detector can be reused at the ad-jacent pixel position.This optimization requires any preprocessing to have a restricted form,such that it takes as input the entire image,and produces as output a new image.The window-by-window preprocessing used in our system cannot be used.A second network is used for precise localization: it is trained to produce a positive response for an exactly centered face,and a negative response for faces which are not centered.It is not trained at all on non-faces.All candidates which produce a posi-tive response from the second network are output as detections.A potential problem in[Vaillant et al., 1994]is that the negative training examples are se-lected manually from a small set of images(indoor scenes,similar to those used for testing the system). It may be possible to make the detectors more robust using the bootstrap technique described here and in [Sung and Poggio,1994].5Conclusions and Future ResearchOur algorithm can detect between78.9%and90.5% of faces in a set of130total images,with an accept-able number of false detections.Depending on the application,the system can be made more or less conservative by varying the arbitration heuristics or thresholds used.The system has been tested on a wide variety of images,with many faces and uncon-strained backgrounds.There are a number of directions for future work. The main limitation of the current system is that it only detects upright faces looking at the camera. Separate versions of the system could be trained for different head orientations,and the results could be combined using arbitration methods similar to those presented here.Other methods of improving system performance in-clude obtaining more positive examples for training, or applying more sophisticated image preprocess-ing and normalization techniques.For instance,theTable2:Comparison of[Sung and Poggio,1994]and Our System on Test Set BMissed Detectdetects rate 10)Networks1and2AND(0)threshold(2,3)overlap elimination3478.1%11)Networks1and2threshold(2,2)overlapelimination AND(2)2087.1%12)Networks1and2threshold(2,2)overlap elimOR(2)threshold(2,1)overlap elimination1192.9%51in1929655131in742175color segmentation method used in[Hunke,1994] for color-based face tracking could be used toﬁlter images.The face detector would then be applied only to portions of the image which contain skin color,which would speed up the algorithm as well as eliminating false detections.One application of this work is in the area of me-dia technology.Every year,improved technology provides cheaper and more efﬁcient ways of storing information.However,automatic high-level classi-ﬁcation of the information content is very limited; this is a bottleneck that prevents media technology from reaching its full potential.The work described above allows a user to make queries of the form “Which scenes in this video contain human faces?”and to have the query answered automatically. AcknowledgementsThe authors would like to thank to Kah-Kay Sung and Dr.Tomaso Poggio(at MIT)and Dr.Woodward Yang(at Harvard)for providing a series of test im-ages and a mug-shot database,respectively.Michael Smith(at CMU)provided some digitized television images for testing purposes.We also thank Eugene Fink,Xue-Mei Wang,Hao-Chi Wong,Tim Rowley, and Kaari Flagstad for comments on drafts of this paper.References[Baluja and Pomerleau,1995]Shumeet Baluja and Dean Pomerleau.Encouraging distributed input reliance in spatially constrained artiﬁcial neuralnetworks:Applications to visual scene analysis and control.Submitted,1995.[Hunke,1994]H.Martin Hunke.Locating and tracking of human faces with neural networks.Master’s thesis,University of Karlsruhe,1994. [Le Cun et al.,1989]Y.Le Cun,B.Boser,J.S.Denker,D.Henderson,R.E.Howard,W.Hub-bard,and L.D.Jackel.Backpropogation ap-plied to handwritten zip code recognition.Neural Computation,1:541–551,1989.[Rowley et al.,1995]Henry A.Rowley,Shumeet Baluja,and Takeo Kanade.Human face detection in visual scenes.CMU-CS-95-158R,Carnegie Mellon University,November1995.Also avail-able at /˜har/faces.html. [Sung and Poggio,1994]Kah-Kay Sung and Tomaso Poggio.Example-based learning for view-based human face detection. A.I.Memo 1521,CBCL Paper112,MIT,December1994. [Umezaki,1995]Tazio Umezaki.Personal com-munication,1995.[Vaillant et al.,1994]R.Vaillant,C.Monrocq,and Y.Le Cun.Original approach for the localisation of objects in images.IEE Proceedings on Vision, Image,and Signal Processing,141(4),August 1994.[Waibel et al.,1989]Alex Waibel,Toshiyuki Hanazawa,Geoffrey Hinton,Kiyohiro Shikano, and Kevin ng.Phoneme recognition using time-delay neural networks.Readings in Speech Recognition,pages393–404,1989.。

Neural Network-Based Face Detection翻译

Neural Network-Based Face Detection 今天读的这篇是基于神经网络的人脸检测。

摘要：1、使用multiple network而不是single network，每个神经网络处理输入图像的一部分2、在收集负样本时采用了a bootstrap algorithm（自举算法）也就是重抽样技术。

（自举法是在1个容量为n的原始样本中重复抽取一系列容量也是n 的随机样本，并保证每次抽样中每一样本观察值被抽取的概率都是1/n(复置抽样)。

这种方法可用来检查样本统计数θ的基本性质，估计θ的标准误和确定一定置信系数下θ的置信区间。

）3、利用了在图像中人脸几乎不重叠这个事实using the fact that faces rarely overlap in images1Introduction:人脸图像可以以人脸图像集的像素强度的概率模型为特征，模型参数由样本图像自动调整。

为人脸检测训练神经网络的难点在于：描述非人脸图像的特征the difficulty in characterizing prototypical（原型的，典型的）“nonface”images.与人脸识别face recognition不同，人脸识别需要辨别不同人脸的种类。

而人脸检测face detection只需要辨别两类：包含人脸的图像images containing faces和不包含人脸的图像images not containing faces.通过有选择性的在训练过程中往训练样本集中添加图像从而避免使用大的非人脸训练集。

这种自举法减少了所需训练集的规模。

使用多个神经网络的仲裁arbitration between multiple networks以及用启发法对结果进行处理heuristics to clean up the results能够有效的提高检测器的性能。

2DESCRIPTION OF THE SYSTEM该人脸检测系统分为两个阶段：stage one:A Neural Network-Based Filter系统的第一个组成部分是一个基于神经网络的滤波器，该滤波器的输入是图像上大小为20*20像素的范围，产生的输出为1或-1，分别代表存在人脸和不存在人脸两种状态。

【毕业论文】基于人工神经网络的人脸识别方法研究

2.5 人脸识别应用领域 .......................................................................................... 18 2.6 现有的人脸识别系统的产品介绍.................................................................... 19 2.7 人脸识别技术所面临的问题 ........................................................................ 20 第三章基于KL变换的人脸特征提取.......................................................................21 3.1 KL算法............................................................................................................. 21 3.2 矩阵分解算法 .................................................................................................. 23
2.1 生物识别技术 .................................................................................................... 7
2.1.1 生物识别的定义 .....................................................................................................7 2.1.2 生物识别技术 ........................................................................................................7

基于PCA和神经网络的人脸识别算法研究

基于PCA和神经网络的人脸识别算法研究作者：唐赫来源：《软件导刊》2013年第06期摘要：在MATLAB环境下，取ORL人脸数据库的部分人脸样本集，基于PCA方法提取人脸特征，形成特征脸空间，然后将每个人脸样本投影到该空间得到一投影系数向量，该投影系数向量在一个低维空间表述了一个人脸样本，这样就得到了训练样本集。

同时将另一部分ORL人脸数据库的人脸作同样处理得到测试样本集。

然后基于最近邻算法进行分类，得到识别率，接下来使用BP神经网络算法进行人脸识别，最后通过基于神经网络算法和最近邻算法进行综合决策，对待识别的人脸进行分类。

关键词：人脸识别；主成分；BP神经网络；最近邻算法中图分类号：TP311文献标识码：A文章编号：1672-7800（2013）006-0033-02作者简介：唐赫（1989-），女，武汉理工大学理学院统计系硕士研究生，研究方向为人脸图像识别、遥感图像、统计预测决策。

0引言特征脸方法就是将人脸的图像域看作是一组随机向量，可以从训练图像中，通过主元分析得到一组特征脸图像，任意给定的人脸图像都可以近似为这组特征脸图像的线性组合，用组合的系数作为人脸的特征向量。

识别过程就是将人脸图像映射到由特征脸组成的子空间上，比较其与已知人脸在特征脸空间中的位置。

经典的特征脸方法是采用基于欧氏距离的最近中心分类器，比较常用的是基于欧氏距离的最近邻。

1算法流程（1）读入人脸库。

每个人取前5张作为训练样本，后5张为测试样本，共40人，则训练样本和测试样本数分别为N=200。

人脸图像为92×112维，按列相连就构成N=10 304维矢量x-j，可视为N维空间中的一个点。

（2）构造平均脸和偏差矩阵。

（3）计算通（4）计算训练样本在特征脸子空间上的投影系数向量，生成训练集的人脸图像主分量allcoor-200×71。

（5）计算测试样本在特征脸子空间上的投影系数向量，生成测试集的人脸图像主分量tcoor-200×71。

基于YOLO多模型交叠桥接策略的人脸检测算法

^mmmm2021年第01期(总第217期）基于YOLO多模型交叠桥接策略的人脸检测算法朱泽群，万秋波,屈金山(三峡大学计算机与信息学院，湖北宜昌443002)摘要:基于深层卷积神经网络的人脸检测算法因其能够较好地克服复杂环境中诸多因素造成的影响，得到了业界广泛关注。

YOLO是一种分类/定位回归式视觉目标检测算法，采用单步检测方式，兼具速度快准确率高的优点，是目前被广泛使用的一种深层全卷积神经网络。

但由于其网络输入尺寸固定，其输出神经元的感受野范围也因此被限定。

将其用于检测尺寸跨度大的人脸目标时，往往会出现无法同时顾及较大或较小人脸目标的问题。

针对该问题，该文提出了一种基于Y O L O多模型交叠桥接策略的大跨度尺寸范围下的人脸检测算法。

首先推算感受野范围，估计出不同输入尺寸的 YOLO网络人脸检测范围；接着，根据各尺寸的检测效果，找到它们的最佳交叠桥接尺寸范围；然后，利用NMS融合各模型检测结果。

在WIDER_FACE数据集的实验结果表明，上述方式可以有效扩展YOLO网络人脸检测的感知范围，提升检测准确率，具有较好的应用前景。

关键词:人脸检测；深度学习；卷积神经网络;YOLO;多尺度中图分类号:TP391 文献标识码:B 文章编号:2096-9759( 2021)01-0044-03Face detection algorithm based on YOLO multi-model Overlapping bridge strategyZhu Zequn, Wan Qiubo, Qu Jinshan(College of Computer and Information Technology, China Three Gorges University, Yichang 443002, China) Abstract:Face detection algorithm based on deep convolutional neural network is widely concerned because it can better over- come the influence of many factors in complex environment, Y OLO is a classification/positioning regressive visual target detec- tion algorithm. It adopts a single-step detection method with the advantages of high speed and accuracy. It is a deep full convolutional neural network widely used at present.However, because the input size of the network is fixed, the receptive field range of t he output neurons is limited. W hen it is used to detect face targets with large size span, it often appears that it cannot take into acco皿t Ae 1虹ge or smdl face targets at the same time.To solve tWs problem, tWs paper proposes a face detection algorithm bas- ed on YOLO multi-model overlapping bridge strategy in large span size range.Firstly, the sensing field range is calculated, and the YOLO network face detection range with different input sizes is estimated.Then, according to the detection effect of each size, find the best overlap bridge size range;Then, NMS is used to fuse the detection results of each model.The experimental results in WIDER_FACE dataset show that the above methods can effectively expand the perceptual range of Y OLO network face detection, improve the detection accuracy, and have a good application prospect.Key words: face detection; deep learning; CNN(Convolutional Neural Network); YOLO; multiscale〇引言随着计算机视觉技术的不断进步，人脸识别系统现已在安全监控、人机交互、犯罪识别、门禁等多种应用领域起到了非常重要的作用，而人脸检测问题作为人脸识别系统中的核心阶段，近几年在计算机目标检测领域更是弓丨起了广泛的关注与研宄。

[论文理解]人脸识别论文总结（一）

[论⽂理解]⼈脸识别论⽂总结（⼀）Face Recognition Papers Review主要两个贡献，⼀是把fc 的权重存到不同卡上去，称为model parallel ，⼆是随机选择negative pair 来近似softmax 的分母(很常规的做法)。

Model Parallel ：FC 分类分配到n 台显卡上，每台显卡分C/n 类，每张卡存权重的⼀部分，计算局部每张卡上的exp 和sumexp ，然后交互计算softmax 。

考虑梯度回传问题，这样做梯度也是parallel 的，不同于数据parallel 。

数据parallel 的话求梯度是需要⽤整个W 才能求W 的梯度的，⽽model paralle 因为有了梯度公式，可知：∇logits i =prob i −onehot i这⼀下明朗了，所以求权重W i 的梯度就有∇W i =X T ∗∇logits i不需要整个W 就可以求部分W 的梯度啦。

作者觉得尽管model parallel 了，但softmax 的分母部分还是⼤啊，于是借鉴常⽤的⽆监督⽅法，随机sample negative pairs ，不需要全部的negative pair 就可以估计出softmax的分母了。

主要提出⼀种动态更新权重的池⼦⽅法，⽤单独⼀个特征⽹络来提取特征作为权重，⽽⾮直接学全连接的权重，然后动态更新这个池⼦，就不需要存储⼤量的权重了，加速了训练、。

⽅法其实很朴素。

前⾯那个⽅法是把权重存到不同的GPU 上去，因此，如果ID 越来越多，我们加卡就可以了，但本⽂的⽅法不需要，也是节约成本的⼀个⽅法。

⽅法⼤致如下：准备两个⽹络，P ⽹络⽤来训练，G ⽹络是P ⽹络的moving avg ，不训练，最开始随机初始化池⼦，记好当前batch 的id ，如果id 在池⼦⾥，训练P ⽹络⽤CE Loss ，和cosine loss ，如果不在⽤cosine loss ，训练⼀轮后更新G ⽹络。

libfacedetection原理

libfacedetection原理libfacedetection是一个基于深度学习的人脸检测库，它的原理是通过神经网络模型来识别和定位图像中的人脸。

本文将对libfacedetection的原理进行详细介绍。

一、libfacedetection的背景和意义随着计算机视觉和人工智能的发展，人脸检测成为了许多应用领域的基础技术，如人脸识别、人脸表情分析、人脸美化等。

而libfacedetection作为一个高效且准确的人脸检测库，可以帮助开发者在各种场景下快速实现人脸检测功能，具有很高的实用价值。

二、libfacedetection的原理介绍libfacedetection的原理主要基于深度学习技术，采用了卷积神经网络（Convolutional Neural Network，CNN）来实现人脸检测。

CNN是一种专门用于处理具有类似网格结构数据的神经网络模型，其结构包含多层卷积层、池化层和全连接层。

具体来说，libfacedetection使用了一个经过预训练的CNN模型，该模型在大规模人脸数据集上进行了训练，学习到了人脸的特征表示。

在进行人脸检测时，libfacedetection将输入的图像通过卷积层进行特征提取，随后通过池化层进行特征降维，最后通过全连接层得到最终的检测结果。

libfacedetection的CNN模型采用了多尺度检测的策略，即在不同的图像尺度下进行人脸检测。

这样可以提高检测算法对不同尺度人脸的适应能力。

另外，libfacedetection还使用了多层级的特征金字塔，通过在不同层次上提取特征，进一步提高了检测算法的准确性。

三、libfacedetection的优势和应用场景相比于传统的人脸检测算法，libfacedetection具有以下几个优势：1. 高效性：libfacedetection采用了深度学习技术，能够在保持高准确性的同时实现较快的检测速度，适用于实时场景。

基于神经网络的多特征轻度认知功能障碍检测模型

第 62 卷第 6 期2023 年11 月Vol.62 No.6Nov.2023中山大学学报（自然科学版）（中英文）ACTA SCIENTIARUM NATURALIUM UNIVERSITATIS SUNYATSENI基于神经网络的多特征轻度认知功能障碍检测模型*王欣1，陈泽森21. 中山大学外国语学院，广东广州 5102752. 中山大学航空航天学院，广东深圳 518107摘要：轻度认知功能障是介于正常衰老和老年痴呆之间的一种中间状态，是老年痴呆诊疗的关键阶段。

因此，针对潜在MCI老年人群进行早期检测和干预，有望延缓语言认知障碍及老年痴呆的发生。

本文利用患者在语言学表现变化明显的特点，提出了一种基于神经网络的多特征轻度认知障碍检测模型。

在提取自然会话中的语言学特征的基础上，融合LDA模型的T-W矩阵与受试者资料等多特征信息，形成TextCNN网络的输入张量，构建基于语言学特征的神经网络检测模型。

该模型在DementiaBank数据集上达到了0.93的准确率、1.00的灵敏度、0.8的特异度和0.9的精度，有效提高了利用自然会话对老年语言认知障碍检测的准确率。

关键词：轻度认知功能障碍；自然会话；神经网络模型；多特征分析；会话分析中图分类号：H030 文献标志码：A 文章编号：2097 - 0137（2023）06 - 0107 - 09A neural network-based multi-feature detection model formild cognitive impairmentWANG Xin1， CHEN Zesen21. School of Foreign Languages， Sun Yat-sen University， Guangzhou 510275， China2. School of Aeronautics and Astronautics， Sun Yat-sen University， Shenzhen 518107， ChinaAbstract：Mild cognitive impairment （MCI） is both an intermediate state between normal aging and Alzheimer's disease and the key stage in the diagnosis of Alzheimer's disease. Therefore， early detec‐tion and treatment for potential elderly can delay the occurrence of dementia. In this study， a neural net‐work-based multi-feature detection model for mild cognitive impairment was proposed， which exploits the characteristics of patients with obvious changes in linguistic performance. The model is based on ex‐tracting the linguistic features in natural speech and integrating the T-W matrix of the LDA model with the subject data and other multi-feature information as the input tensor of the TextCNN network. It achieved an accuracy of 0.93， a sensitivity of 1.00， a specificity of 0.8， and a precision of 0.9 on the DementiaBank dataset， which effectively improved the accuracy of cognitive impairment detection in the elderly by using natural speech.Key words：mild cognitive impairment； natural speech； neural network model； multi-feature detec‐tion； speech analysisDOI：10.13471/ki.acta.snus.2023B049*收稿日期：2023 − 07 − 18 录用日期：2023 − 07 − 30 网络首发日期：2023 − 09 − 21基金项目：教育部人文社会科学基金（22YJCZH179）；中国科协科技智库青年人才计划（20220615ZZ07110400）；中央高校基本科研业务费重点培育项目（23ptpy32）作者简介：王欣（1991年生），女；研究方向：应用语言学；E-mail：******************第 62 卷中山大学学报（自然科学版）（中英文）轻度认知障碍（MCI，mild cognitive impair‐ment）是一种神经系统慢性退行性疾病，也是阿尔茨海默病（AD，Alzheimer's disease）的早期关键阶段。

基于YOLOv3的人脸关键点检测

^mmmm2021年第01期(总第217期）基于YOLOv3的人脸关键点检测屈金山，朱泽群,万秋波(三峡大学计算机与信息学院，湖北宜昌443002)摘要：深度学习中神经网络强大的特征提取能力，使非约束场景下的人脸检测不再困难，于是人脸关键点的检测逐渐成为人脸检测的关注点，但目前为止较少算法具备对人脸关键点的检测能力。

Y O L O v3作为精度和速度均表现优异的算法，同样不具备关键点检测的能力。

因此，文章提出基于Y O L O v3的人脸关键点检测算法，该算法对Y O L O v3改进，设计关键点损失函数，实现对人脸关键点的定位，最终实现Y O L O v3在人脸检测中同时输出人脸包围框和人脸关键点。

实验表明，提出的方法在Y O L O v3上成功实现了对人脸矩形包围框和人脸关键点的同时输出。

关键词:人脸检测；深度学习；Y O L O v3;关键点检测;损失函数中图分类号:TP391 文献标识码:B文章编号=2096-9759(2021)01-0055-04F a ceLandmarkDetectionBasedOn Y O LOv3Qu Jinshan，Zu Zequn，Wan Qiubo(School of Computer and Information science,China Three Gorges University,Yichang Hubei 443002) Abstract: The powerful feature extraction ability of neural network in deep learning makes face detection in uncon-strained scenes no longer diffic ult，so the detection of face key points has gradually become the focus of face detection，but so fa r，few algorithms have the ability to detect face key points.As an algorithm with excellent accuracy and speed,yolov3 also does not have the ability of key point detection.Therefore，this paper proposes a face key point detection al-gorithm based on yolov3.The algorithm improves yolov3,designs the key point loss function,realizes the location of the face key points,and finally realizes the simultaneous output of face bounding box and face key points in yolov3 face de-tection.Experimental results show that the proposed method can successfully output the rectangular bounding box and key points of human face on yolov3.Key words: face detection;deep learning;Y O L O v3; face landmark;loss function〇引言人脸检测是机器视觉领域被深入研宄的经典问题，早期人脸检测作为人脸识别的一部分，待检测的人脸通常为受到约束的特征明显的人脸，具有清晰的五官特征以及较小的尺度变化。

retinaface原理

retinaface原理
RetinaFace原理：深度学习技术在人脸检测中的应用
RetinaFace是一种基于深度学习技术的人脸检测算法，它可以在复杂的场景中准确地检测出人脸，并且能够识别出人脸的五官和面部特征。

RetinaFace的原理是通过卷积神经网络（CNN）来实现的，它可以对图像进行高效的特征提取和分类，从而实现对人脸的检测和识别。

RetinaFace的核心是一个多任务级联卷积神经网络（MTCNN），它由三个级联的卷积神经网络组成，分别是Proposal Network（P-Net）、Refine Network（R-Net）和Output Network（O-Net）。

P-Net用于生成候选框，R-Net用于对候选框进行筛选和优化，O-Net 用于对最终的人脸框进行分类和回归。

RetinaFace的优点在于它可以同时检测多个人脸，并且可以检测出不同大小和角度的人脸。

此外，RetinaFace还可以识别出人脸的五官和面部特征，包括眼睛、鼻子、嘴巴等，从而实现更加精准的人脸识别和分析。

RetinaFace的应用非常广泛，它可以用于人脸识别、人脸跟踪、人脸表情识别、人脸属性分析等领域。

在安防领域，RetinaFace可以用于监控视频中的人脸检测和识别，从而实现对不法分子的追踪和抓捕。

在医疗领域，RetinaFace可以用于人脸识别和面部特征分析，
从而实现对疾病的诊断和治疗。

RetinaFace是一种非常先进的人脸检测算法，它利用深度学习技术实现了对人脸的高效检测和识别，具有广泛的应用前景。

随着深度学习技术的不断发展和完善，RetinaFace将会在更多的领域得到应用，并为人们的生活带来更多的便利和安全。

基于卷积神经网络的人脸检测算法研究

基于卷积神经网络的人脸检测算法研究作者：刘天保来源：《软件导刊》2020年第10期摘要：近年来，随着深度学习的迅猛发展，人脸检测算法准确度已有很大提升。

模型越复杂，检测速度越慢，设计一种准确度与速度兼顾的人脸检测模型尤为必要。

基于FaceBoxes 人脸检测算法框架，提出一种基于深层卷积主干网络的改进方法，并在人脸检测基准数据集中进行测试实验。

其在FDDB数据集上的实验结果显示，检测正确率达95%，比传统方法提高1.67%。

该算法在保证实时性的同时提升了检测准确率，可应用于追求更高准确率的人脸检测系统。

关键词：人脸检测; 深度学习; 卷积神经网络DOI：10. 11907/rjdk. 201219中图分类号：TP312 文献标识码：A 文章编号：1672-7800（2020）010-0066-05Abstract： Thanks to the rapid development of deep learning in recent years， the accuracy of face detection algorithm has been greatly improved compared with the earlier algorithms. However，the more complex the model detection speed will be slower， so the design of a face detection model with both accuracy and speed has become a major topic in this field. Based on FaceBoxes， a face detection algorithm framework， this paper proposes an improved method of deep convolutional backbone network， and conducts test experiments in the face detection benchmark data set. The experimental results on the FDDB data set showed that the detection accuracy reached 95%， which was 1.67% higher than the traditional method. The algorithm in this paper not only guarantees real-time performance but also improves the accuracy of detection， which can be used in more accurate face detection systems.Key Words： face detection; deep learning; convolutional neural network0 引言目前，人脸检测技术已广泛应用于人们的日常生活中，如机器人[1]、自动驾驶[2]、逃犯追捕[3]、考勤打卡[4]、门禁系统[5]等都用到了人脸检测技术。

基于语义分割的人脸篡改检测研究

摘要如今，图像生成技术获得了快速发展，假脸生成技术也应运而生。

其中具有代表的技术有Deepfake、Face2Face[1]、NeuralTextures[2]等等。

人们可以利用这些模型完成图像或视频中人脸的互换，甚至是生成一个不存在的人脸。

这些技术主要包含三个步骤——人脸定位、人脸转换和人脸拼接。

目前，人脸定位技术已经发展地十分成熟，一般的人脸定位算法主要是找到人脸上的特征点，比如鼻子、嘴巴、下巴以及眉毛等，通过抽取这些特征可以知道人脸的器官分布，完成对人脸的定位。

例如，通过调用dlib库以及OpenCV的包来进提取这些特征点，也可以利用深度神经网络如CNN模型来实现定位。

人脸转换主要包括对抗生成模型(GAN)和变分自编码器(V AE)等生成模型，通过训练对应的编码解码器来完成两个人之间的人脸互换。

例如，V AE通过无监督的方式将人脸图像压缩成向量，再将这个向量恢复成人脸图像，如果对人A的人脸图像进行编码，再用人B的解码器对编码后的图像进行解码，就可以使B的人脸图像中人脸表情变成A的表情。

而GAN则是抽取人脸的特征点，生成器用这些特征生成目标人脸图像，通过训练判别器使得判别器不能区分生成人脸和真实人脸，完成人脸的生成。

Deepfake等深度学习技术愈演愈烈的同时，也带来了巨大的隐患。

不得不承认换脸是人们滥用深度学习的后果。

在此之前，可以在社交媒体上大胆的发布照片，例如朋友圈，微博，Facebook等，并且人们对网络新闻中的视频和图像也是较信服的。

而自从这些换脸技术开源之后，可以发现有人通过换脸技术处理并发布一些虚假视频，更有甚者，利用收集到的人脸图像生成他人人脸的色情影片。

因此，对抗Deepfake技术是同样重要。

目前已有很多文章聚焦于反Deepfake技术领域，主流的检测方法是在真实图像和和伪造图像中训练一个二分类模型，来判别测试图像是否是生成的。

这样做的缺点是模型对数据集存在依赖性，分类器无法判别对于用不同生成模型生成的数据集。

基于卷积神经网络的口罩遮挡条件下人脸识别技术研究

总777期第七期2022年4月河南科技Henan Science and Technology信息技术基于卷积神经网络的口罩遮挡条件下人脸识别技术研究颜继东李艳生（湖北师范大学物理与电子科学学院，湖北黄石435002）摘要：为了解决戴口罩人脸识别率不高的问题，开展基于卷积神经网络的研究。

本研究所用的数据集均为戴口罩的人脸图像，总共1016张图片，其中测试样本为305张，训练样本为711张。

本研究采用对比试验的方法，在初始卷积神经网络模型结构不变的情况下，修改每一个卷积层的卷积核数量，从后往前进行对比，得出本研究最佳模型，人脸检测准确率约为99.74%。

通过试验可以看出，卷积神经网络对戴口罩的人脸有很好的识别能力。

关键词：人脸识别；卷积神经网络；深度学习中图分类号：TP391文献标志码：A文章编号：1003-5168（2022）7-0010-05 DOI:10.19968/ki.hnkj.1003-5168.2022.07.002Research on Face Recognition Technology Under Mask Occlusion Basedon Convolutional Neural NetworkYAN Jidong LI Yansheng(College of Physics and Electronics,Hubei Normal University,Huangshi435002,China)Abstract:In order to solve the problem of low face recognition rate with mask in the epidemic environ⁃ment.This study is the convolution neural network based study of the problem.The data sets used in this experiment all use mask faces,with a total of1016pictures,including305test samples and711training samples.The research process adopts the method of comparative experiment.Under the condition that the structure of the initial convolution neural network model remains unchanged,modify the number of convolution cores of each convolution layer,compare it from back to front,and get the best model in this experiment.The accuracy of face detection is about99.74%.Through this experiment,it can be seen that the convolutional neural network has a good ability to recognize the face wearing mask. Keywords:face recognition;convolutional neural network;deep learning0引言神经网络是通过模仿人类大脑神经元之间的生物信号传递机制而诞生的计算机交叉科学［1~2］。

基于ResNet50的人脸识别模型

9科技资讯 SCIENCE & TECHNOLOGY INFORMATION①基金项目：南京工程学院本科生科技创新训练项目《基于NVIDIA JETSON NANO的人脸识别系统模型》（项目编号：TB202004017）。

作者简介：谢鑫（1998—），男，汉族，山东临沂人，本科在读，研究方向为电气工程与智能控制。

DOI：10.16661/ki.1672-3791.2004-4912-0634基于ResNet50的人脸识别模型①谢鑫(南京工程学院电力工程学院江苏南京 211167)摘要：该文采用卷积神经网络（Convolutional Neural Networks, CNN ）进行人脸识别模型开发，现有深度卷积神经网络的模型，提出了改进的Faster R-CNN目标检测模型与ResNet50结合的方法进行识别。

ResNet50网络大小适中，对算力要求不高，故可以采用训练耗时较长的杜鹃算法（Cuckoo Search，CS ）对原有的网络训练方法进行一定变更，在CelebA上进行训练与预测。

与未改进的ResNet50相比，其训练时间有所增加但预测准确率上升1.7%。

关键词：ResNet50 Faster r c nn Cuckoo Search 人脸识别中图分类号：TP391.41 文献标识码：A 文章编号：1672-3791(2020)12(b)-0009-03Face Recognition Model Based on ResNet50XIE Xin(School of Electrical Engineering, Nanjing Institute of Technology, Nanjing, Jiangsu Province, 211167 China)Abstract: This paper uses Convolutional Neural Networks (CNN) for face recognition model development. For existing deep convolutional neural network models, an improved Faster R-CNN target detection model combined with ResNet50 is proposed for recognition. The ResNet50 network is of moderate size and does not require high computing power. Therefore, the Cuckoo Search (CS) algorithm, which takes a long time to train, can be used to make certain changes to the original network training method, and train and predict on CelebA. Compared with the unimproved ResNet50, its training time has increased but the prediction accuracy has increased by 1.7%.Key Words: ResNet50; Faster r c nn; Cuckoo Search; Face recognitionYann LeCun于1989年将CNN用于手写数字识别, 近年来CNN也被广泛用于人脸识别领域[1]。

基于肤色分割和改进Gabor滤波相结合的人脸检测

基于肤色分割和改进Gabor滤波相结合的人脸检测张国伟;于金霞;陈现查【摘要】针对人脸检测问题,提出了一种基于肤色分割和改进Gabor滤波相结合的方法;该方法首先基于YCbCr空间对图像中的背景区域和人脸肤色进行分割,以消除大量的背景区域提高运算速度;接着利用改进的Gabor滤波对提取出来的人脸肤色区域进行卷积得到人脸的特征向量,并和通过训练样本获得的特征向量进行比较以验证是否为人脸;最后通过实验分析,验证了所提方法能够在保证检测精度的基础上有效提高运算速度.【期刊名称】《计算机测量与控制》【年(卷),期】2010(018)001【总页数】4页(P136-138,141)【关键词】人脸检测;改进Gabor滤波;肤色分割【作者】张国伟;于金霞;陈现查【作者单位】河南理工大学计算机科学与技术学院,河南,焦作,454003;河南理工大学计算机科学与技术学院,河南,焦作,454003;江苏省图像处理与图像通信重点实验室,江苏,南京,210003;河南理工大学计算机科学与技术学院,河南,焦作,454003【正文语种】中文【中图分类】TP240 引言近年来,人脸检测已经成为模式识别的研究热点之一。

人脸检测是实现人脸自动识别的第一步,其实质上是实现人脸与背景(非人脸)分离,从各种不同的背景环境中确定人脸的存在,并且确定人脸的数目和位置。

目前,人脸检测的方法可以分为4类[1-4]：基于知识的方法、特征不变量方法、模板匹配方法和基于表象的方法。

在基于表象的人脸检测方法,最经典的是特征脸方法[4](或PCA方法)。

现有的方法大多是基于图像对每个窗口进行复杂计算以确定其是否为人脸[5-6]。

实际上,存在人脸的窗口数量远远小于不存在人脸的窗口数量,并且很多背景区域和人脸的差别很大。

因而,可通过肤色来快速过滤掉大量的背景区域。

对于较复杂的背景区域,可以使用训练好的滤波器作进一步的特征提取和验证。

综上分析,本文提出了一种基于肤色分割和改进Gabor滤波相结合的人脸检测方法：首先基于YCbCr空间对图像中的背景区域和人脸肤色进行分割,以消除大量的背景区域提高运算速度;接着利用改进的Gabor滤波对提取出的人脸肤色区域进行卷积得到人脸的特征向量,并和通过训练样本获得的特征向量进行比较以验证是否为人脸,最后通过实验分析,验证了所提方法能够在保证检测精度的基础上有效提高运算速度。