Haar-like人脸检测论文资源-CSDN文库

共2个文件

pdf：2个

人脸识别

论文

Adaboost

需积分: 31 142 浏览量 2014-12-12 12:27:12 上传评论 2 收藏 438KB ZIP 举报

资源推荐

资源详情

资源评论

收起资源包目录

归档.zip （2个子文件）

Rapid_Object_Detection_using_a_Boosted_Cascade_of_Simple_Features.pdf 195KB

Robust Real-Time Face Detection.pdf 328KB

International Journal of Computer Vision 57(2), 137–154, 2004

 2004 Kluwer Academic Publishers. Manufactured in The Netherlands.

Robust Real-Time Face Detection

PAUL VIOLA

Microsoft Research, One Microsoft Way, Redmond, WA 98052, USA

viola@microsoft.com

MICHAEL J. JONES

Mitsubishi Electric Research Laboratory, 201 Broadway, Cambridge, MA 02139, USA

mjones@merl.com

Received September 10, 2001; Revised July 10, 2003; Accepted July 11, 2003

Abstract. This paper describes a face detection framework that is capable of processing images extremely rapidly

while achieving high detection rates. There are three key contributions. The ﬁrst is the introduction of a new

image representation called the “Integral Image” which allows the features used by our detector to be computed

very quickly. The second is a simple and efﬁcient classiﬁer which is built using the AdaBoost learning algo-

rithm (Freund and Schapire, 1995) to select a small number of critical visual features from a very large set of

potential features. The third contribution is a method for combining classiﬁers in a “cascade” which allows back-

ground regions of the image to be quickly discarded while spending more computation on promising face-like

regions. A set of experiments in the domain of face detection is presented. The system yields face detection perfor-

mance comparable to the best previous systems (Sung and Poggio, 1998; Rowley et al., 1998; Schneiderman and

Kanade, 2000; Roth et al., 2000). Implemented on a conventional desktop, face detection proceeds at 15 frames per

second.

Keywords: face detection, boosting, human sensing

1. Introduction

This paper brings together new algorithms and insights

to construct a framework for robust and extremely rapid

visual detection. Toward this end we have constructed

a frontal face detection system which achieves detec-

tion and false positive rates which are equivalent to

the best published results (Sung and Poggio, 1998;

Rowley et al., 1998; Osuna et al., 1997a; Schneiderman

and Kanade, 2000; Roth et al., 2000). This face detec-

tion system is most clearly distinguished from previ-

ous approaches in its ability to detect faces extremely

rapidly. Operating on 384 by 288 pixel images, faces

are detected at 15 frames per second on a conventional

700 MHz Intel Pentium III. In other face detection

systems, auxiliary information, such as image differ-

ences in video sequences, or pixel color in color im-

ages, have been used to achieve high frame rates. Our

system achieves high frame rates working only with

the information present in a single grey scale image.

These alternative sources of information can also be in-

tegrated with our system to achieve even higher frame

rates.

There are three main contributions of our face detec-

tion framework. We will introduce each of these ideas

brieﬂy below and then describe them in detail in sub-

sequent sections.

The ﬁrst contribution of this paper is a new image

representation called an integral image that allows for

very fast feature evaluation. Motivated in part by the

work of Papageorgiou et al. (1998) our detection sys-

tem does not work directly with image intensities. Like

138 Viola and Jones

these authors we use a set of features which are rem-

iniscent of Haar Basis functions (though we will also

use related ﬁlters which are more complex than Haar

ﬁlters). In order to compute these features very rapidly

at many scales we introduce the integral image repre-

sentation for images (the integral image is very similar

to the summed area table used in computer graphics

(Crow, 1984) for texture mapping). The integral im-

age can be computed from an image using a few op-

erations per pixel. Once computed, any one of these

Haar-like features can be computed at any scale or lo-

cation in constant time.

The second contribution of this paper is a simple

and efﬁcient classiﬁer that is built by selecting a small

number of important features from a huge library of po-

tential features using AdaBoost (Freund and Schapire,

1995). Within any image sub-window the total num-

ber of Haar-like features is very large, far larger than

the number of pixels. In order to ensure fast classiﬁ-

cation, the learning process must exclude a large ma-

jority of the available features, and focus on a small

set of critical features. Motivated by the work of Tieu

and Viola (2000) feature selection is achieved using

the AdaBoost learning algorithm by constraining each

weak classiﬁer to depend on only a single feature. As a

result each stage of the boosting process, which selects

anew weak classiﬁer, can be viewed as a feature selec-

tion process. AdaBoost provides an effective learning

algorithm and strong bounds on generalization perfor-

mance (Schapire et al., 1998).

The third major contribution of this paper is a method

for combining successively more complex classiﬁers

in a cascade structure which dramatically increases the

speed of the detector by focusing attention on promis-

ing regions of the image. The notion behind focus

of attention approaches is that it is often possible to

rapidly determine where in an image a face might oc-

cur (Tsotsos et al., 1995; Itti et al., 1998; Amit and

Geman, 1999; Fleuret and Geman, 2001). More com-

plex processing is reserved only for these promising

regions. The key measure of such an approach is the

“false negative” rate of the attentional process. It must

be the case that all, or almost all, face instances are

selected by the attentional ﬁlter.

We will describe a process for training an extremely

simple and efﬁcient classiﬁer which can be used as a

“supervised” focus of attention operator.

Aface de-

tection attentional operator can be learned which will

ﬁlter out over 50% of the image while preserving 99%

of the faces (as evaluated over a large dataset). This

ﬁlter is exceedingly efﬁcient; it can be evaluated in 20

simple operations per location/scale (approximately 60

microprocessor instructions).

Those sub-windows which are not rejected by the

initial classiﬁer are processed by a sequence of classi-

ﬁers, each slightly more complex than the last. If any

classiﬁer rejects the sub-window, no further processing

is performed. The structure of the cascaded detection

process is essentially that of a degenerate decision tree,

and as such is related to the work of Fleuret and Geman

(2001) and Amit and Geman (1999).

The complete face detection cascade has 38 classi-

ﬁers, which total over 80,000 operations. Nevertheless

the cascade structure results in extremely rapid average

detection times. On a difﬁcult dataset, containing 507

faces and 75 million sub-windows, faces are detected

using an average of 270 microprocessor instructions

per sub-window. In comparison, this system is about

15 times faster than an implementation of the detection

system constructed by Rowley et al. (1998).

An extremely fast face detector will have broad prac-

tical applications. These include user interfaces, im-

age databases, and teleconferencing. This increase in

speed will enable real-time face detection applications

on systems where they were previously infeasible. In

applications where rapid frame-rates are not necessary,

our system will allow for signiﬁcant additional post-

processing and analysis. In addition our system can be

implemented on a wide range of small low power de-

vices, including hand-helds and embedded processors.

In our lab we have implemented this face detector on a

low power 200 mips Strong Arm processor which lacks

ﬂoating point hardware and have achieved detection at

two frames per second.

1.1. Overview

The remaining sections of the paper will discuss the

implementation of the detector, related theory, and ex-

periments. Section 2 will detail the form of the features

as well as a new scheme for computing them rapidly.

Section 3 will discuss the method in which these fea-

tures are combined to form a classiﬁer. The machine

learning method used, a application of AdaBoost, also

acts as a feature selection mechanism. While the classi-

ﬁers that are constructed in this way have good compu-

tational and classiﬁcation performance, they are far too

slow for a real-time classiﬁer. Section 4 will describe a

method for constructing a cascade of classiﬁers which

Robust Real-Time Face Detection 139

together yield an extremely reliable and efﬁcient face

detector. Section 5 will describe a number of experi-

mental results, including a detailed description of our

experimental methodology. Finally Section 6 contains

a discussion of this system and its relationship to re-

lated systems.

2. Features

Our face detection procedure classiﬁes images based

on the value of simple features. There are many moti-

vations for using features rather than the pixels directly.

The most common reason is that features can act to en-

code ad-hoc domain knowledge that is difﬁcult to learn

using a ﬁnite quantity of training data. For this system

there is also a second critical motivation for features:

the feature-based system operates much faster than a

pixel-based system.

The simple features used are reminiscent of Haar

basis functions which have been used by Papageorgiou

et al. (1998). More speciﬁcally, we use three kinds of

features. The value of a two-rectangle feature is the

difference between the sum of the pixels within two

rectangular regions. The regions have the same size

and shape and are horizontally or vertically adjacent

(see Fig. 1). A three-rectangle feature computes the

sum within two outside rectangles subtracted from the

sum in a center rectangle. Finally a four-rectangle fea-

ture computes the difference between diagonal pairs of

rectangles.

Given that the base resolution of the detector is

24 × 24, the exhaustive set of rectangle features is

Figure 1. Example rectangle features shown relative to the enclos-

ing detection window. The sum of the pixels which lie within the

white rectangles are subtracted from the sum of pixels in the grey

rectangles. Two-rectangle features are shown in (A) and (B). Figure

quite large, 160,000. Note that unlike the Haar basis,

the set of rectangle features is overcomplete.

2.1. Integral Image

Rectangle features can be computed very rapidly using

an intermediate representation for the image which we

call the integral image.

The integral image at location

x, y contains the sum of the pixels above and to the left

of x, y, inclusive:

ii(x, y) =



≤x ,y



≤y

i(x



, y



where ii(x, y)isthe integral image and i(x, y)isthe

original image (see Fig. 2). Using the following pair of

recurrences:

s(x, y) = s(x, y − 1) + i(x, y) (1)

ii(x, y) = ii(x − 1, y) + s(x, y) (2)

(where s(x, y)isthe cumulative row sum, s(x, −1) =

0, and ii(−1, y) = 0) the integral image can be com-

puted in one pass over the original image.

Using the integral image any rectangular sum can be

computed in four array references (see Fig. 3). Clearly

the difference between two rectangular sums can be

computed in eight references. Since the two-rectangle

features deﬁned above involve adjacent rectangular

sums they can be computed in six array references,

eight in the case of the three-rectangle features, and

nine for four-rectangle features.

One alternative motivation for the integral im-

age comes from the “boxlets” work of Simard et al.

Figure 2. The value of the integral image at point (x, y)isthe sum

of all the pixels above and to the left.

140 Viola and Jones

Figure 3. The sum of the pixels within rectangle D can be computed

with four array references. The value of the integral image at location

1isthe sum of the pixels in rectangle A. The value at location 2 is

A + B,atlocation 3 is A + C , and at location 4 is A + B + C + D.

The sum within D can be computed as 4 + 1 − (2 + 3).

(1999). The authors point out that in the case of linear

operations (e.g. f · g), any invertible linear operation

can be applied to f or g if its inverse is applied to the

result. For example in the case of convolution, if the

derivative operator is applied both to the image and the

kernel the result must then be double integrated:

f ∗ g =



( f



∗ g



The authors go on to show that convolution can be

signiﬁcantly accelerated if the derivatives of f and g

are sparse (or can be made so). A similar insight is that

an invertible linear operation can be applied to f if its

inverse is applied to g:

( f



) ∗







= f ∗ g.

Viewed in this framework computation of the rect-

angle sum can be expressed as a dot product, i ·r , where

i is the image and r is the box car image (with value

1 within the rectangle of interest and 0 outside). This

operation can be rewritten

i · r =







· r



The integral image is in fact the double integral of the

image (ﬁrst along rows and then along columns). The

second derivative of the rectangle (ﬁrst in row and then

in column) yields four delta functions at the corners of

the rectangle. Evaluation of the second dot product is

accomplished with four array accesses.

2.2. Feature Discussion

Rectangle features are somewhat primitive when

compared with alternatives such as steerable ﬁlters

(Freeman and Adelson, 1991; Greenspan et al., 1994).

Steerable ﬁlters, and their relatives, are excellent for the

detailed analysis of boundaries, image compression,

and texture analysis. While rectangle features are also

sensitive to the presence of edges, bars, and other sim-

ple image structure, they are quite coarse. Unlike steer-

able ﬁlters, the only orientations available are vertical,

horizontal and diagonal. Since orthogonality is not cen-

tral to this feature set, we choose to generate a very

large and varied set of rectangle features. Typically the

representation is about 400 times overcomplete. This

overcomplete set provides features of arbitrary aspect

ratio and of ﬁnely sampled location. Empirically it ap-

pears as though the set of rectangle features provide

a rich image representation which supports effective

learning. The extreme computational efﬁciency of rect-

angle features provides ample compensation for their

limitations.

In order to appreciate the computational advantage

of the integral image technique, consider a more con-

ventional approach in which a pyramid of images is

computed. Like most face detection systems, our de-

tector scans the input at many scales; starting at the

base scale in which faces are detected at a size of

24 × 24 pixels, a 384 by 288 pixel image is scanned

at 12 scales each a factor of 1.25 larger than the last.

The conventional approach is to compute a pyramid of

12 images, each 1.25 times smaller than the previous

image. A ﬁxed scale detector is then scanned across

each of these images. Computation of the pyramid,

while straightforward, requires signiﬁcant time. Imple-

mented efﬁciently on conventional hardware (using bi-

linear interpolation to scale each level of the pyramid) it

takes around .05 seconds to compute a 12 level pyramid

of this size (on an Intel PIII 700 MHz processor).

In contrast we have deﬁned a meaningful set of rect-

angle features, which have the property that a single

feature can be evaluated at any scale and location in a

few operations. We will show in Section 4 that effec-

tive face detectors can be constructed with as few as two

rectangle features. Given the computational efﬁciency

of these features, the face detection process can be com-

pleted for an entire image at every scale at 15 frames per

Robust Real-Time Face Detection 141

second, about the same time required to evaluate the 12

level image pyramid alone. Any procedure which re-

quires a pyramid of this type will necessarily run slower

than our detector.

3. Learning Classiﬁcation Functions

Given a feature set and a training set of positive and

negative images, any number of machine learning ap-

proaches could be used to learn a classiﬁcation func-

tion. Sung and Poggio use a mixture of Gaussian model

(Sung and Poggio, 1998). Rowley et al. (1998) use a

small set of simple image features and a neural net-

work. Osuna et al. (1997b) used a support vector ma-

chine. More recently Roth et al. (2000) have proposed

anew and unusual image representation and have used

the Winnow learning procedure.

Recall that there are 160,000 rectangle features as-

sociated with each image sub-window, a number far

larger than the number of pixels. Even though each

feature can be computed very efﬁciently, computing

the complete set is prohibitively expensive. Our hy-

pothesis, which is borne out by experiment, is that a

very small number of these features can be combined

to form an effective classiﬁer. The main challenge is to

ﬁnd these features.

In our system a variant of AdaBoost is used both

to select the features and to train the classiﬁer (Freund

and Schapire, 1995). In its original form, the AdaBoost

learning algorithm is used to boost the classiﬁcation

performance of a simple learning algorithm (e.g., it

might be used to boost the performance of a simple per-

ceptron). It does this by combining a collection of weak

classiﬁcation functions to form a stronger classiﬁer. In

the language of boosting the simple learning algorithm

is called a weak learner. So, for example the percep-

tron learning algorithm searches over the set of possible

perceptrons and returns the perceptron with the lowest

classiﬁcation error. The learner is called weak because

we do not expect even the best classiﬁcation function to

classify the training data well (i.e. for a given problem

the best perceptron may only classify the training data

correctly 51% of the time). In order for the weak learner

to be boosted, it is called upon to solve a sequence of

learning problems. After the ﬁrst round of learning, the

examples are re-weighted in order to emphasize those

which were incorrectly classiﬁed by the previous weak

classiﬁer. The ﬁnal strong classiﬁer takes the form of a

perceptron, a weighted combination of weak classiﬁers

followed by a threshold.

The formal guarantees provided by the AdaBoost

learning procedure are quite strong. Freund and

Schapire proved that the training error of the strong

classiﬁer approaches zero exponentially in the number

of rounds. More importantly a number of results

were later proved about generalization performance

(Schapire et al., 1997). The key insight is that gen-

eralization performance is related to the margin of the

examples, and that AdaBoost achieves large margins

rapidly.

The conventional AdaBoost procedure can be eas-

ily interpreted as a greedy feature selection process.

Consider the general problem of boosting, in which a

large set of classiﬁcation functions are combined using

a weighted majority vote. The challenge is to associate

a large weight with each good classiﬁcation function

and a smaller weight with poor functions. AdaBoost is

an aggressive mechanism for selecting a small set of

good classiﬁcation functions which nevertheless have

signiﬁcant variety. Drawing an analogy between weak

classiﬁers and features, AdaBoost is an effective pro-

cedure for searching out a small number of good “fea-

tures” which nevertheless have signiﬁcant variety.

One practical method for completing this analogy is

to restrict the weak learner to the set of classiﬁcation

functions each of which depend on a single feature.

In support of this goal, the weak learning algorithm is

designed to select the single rectangle feature which

best separates the positive and negative examples (this

is similar to the approach of Tieu and Viola (2000) in

the domain of image database retrieval). For each fea-

ture, the weak learner determines the optimal threshold

classiﬁcation function, such that the minimum num-

ber of examples are misclassiﬁed. A weak classiﬁer

(h(x, f, p,θ)) thus consists of a feature ( f ), a thresh-

old (θ) and a polarity (p) indicating the direction of the

inequality:

h(x, f, p,θ) =



1ifpf(x) < pθ

0 otherwise

Here x isa24× 24 pixel sub-window of an image.

In practice no single feature can perform the classiﬁ-

cation task with low error. Features which are selected

early in the process yield error rates between 0.1 and

0.3. Features selected in later rounds, as the task be-

comes more difﬁcult, yield error rates between 0.4 and

0.5. Table 1 shows the learning algorithm.

The weak classiﬁers that we use (thresholded single

features) can be viewed as single node decision trees.

评论收藏

内容反馈

非主流科学家

粉丝: 1213
资源: 9

Haar-like人脸检测论文

论文研究-一种结合肤色及类人脸特征的人脸检测.pdf

Haar-Like特征和AdaBoost论文学习的PPT

论文研究-一种快速高效的人脸检测方法.pdf

人脸检测的论文资源.参考资料，可以很有用的

研究论文-视频流下的人脸检测与跟踪.pdf

论文研究-AdaBoost算法在人脸检测中的应用 .pdf

论文研究-基于协方差特征的裁剪AdaBoost算法.pdf

论文研究-一种快速的自动人眼定位方法.pdf

视频流下的人脸检测与跟踪 (2011年)

人脸识别特征提取算法研究

论文研究-基于面部结构的表情识别.pdf

Academic+Phrasebank+2021+Edition+_中英文对照.pdf

YOLO v1 ~ YOLO v5 论文解读和实现细节

基于python的超市管理系统的设计与实现毕业论文+项目文档源码

1000套计算机毕业设计带源码

数模国赛word模板.zip

2021年国赛A题（FAST主动反射面形状调节）论文+代码材料.zip

2023高教社数学建模C题 - 蔬菜类商品的自动定价与补货决策【数据处理详细代码】

Python大作业（包含论文）——可打包的双人五子棋程序

基于高校校园网的网络规划设计与实现-以锦城学院为例-kaic.docx

YOLOv9论文，2024.02发布

软考 系统分析师论文 范文

elsevier投稿模板Latex压缩包

ChatGPT4.0中文版论文

边缘计算_万物互联时代新型计算模型_施巍松.pdf

yolov综述论文，v1到v8的详细深入对比剖析

萌宠领养平台微信小程序的设计与实现毕业设计答辩PPT

python地铁客流量分析平台_python毕业设计_爬虫可视化_论文_python_毕业论文

电工杯历年优秀论文 .zip

最新资源

软考系统分析师论文范文