《模式识别第四版》：深度探讨有监督与无监督算法

5星 · 超过95%的资源需积分: 10 164 浏览量更新于2024-07-25 1 收藏 13.12MB PDF 举报

《模式识别——第四版》是一本由学术出版社 Academic Press 发行的权威著作，隶属于跨国科技出版巨头 Elsevier 的一部分。该书作为机器学习领域的经典教材，全面而深入地探讨了模式识别的核心理论和技术。作者通过对有监督学习和无监督学习这两种主要的学习范式进行详细的比较，帮助读者理解并掌握现代模式识别算法的基础和应用。在本书中，读者将了解到各种监督学习方法，如决策树、支持向量机（SVM）、神经网络等，它们如何通过训练数据集学习并建立预测模型，以解决分类和回归问题。同时，无监督学习部分则涉及聚类、降维和关联规则学习等内容，这些技术在数据挖掘和模式发现任务中发挥着关键作用，尤其是在缺乏明确标签的情况下。此外，书中还可能涵盖了模式识别中的特征选择、模型评估和优化、以及深度学习等前沿技术，这些都是提高模式识别系统性能的重要组成部分。作者以清晰的阐述和丰富的实例，确保读者能够掌握从理论到实践的各个环节。版权信息部分强调了书籍的知识产权保护，未经Elsevier书面许可，任何形式的复制或传播都需得到授权。图书已向美国图书馆协会和大英图书馆提交申请，并提供了相应的索引和版权登记号，便于读者查找和获取更多信息。《模式识别——第四版》是一本不可或缺的参考资料，对于希望在这个领域深造的学生、研究者以及工程师来说，它提供了一个全面且深入的平台，帮助他们理解和开发高效、实用的模式识别解决方案。通过阅读这本书，读者不仅能掌握理论知识，还能培养实际操作和解决问题的能力。

“03-Ch01-SA272” 17/9/2008 page 2

2 CHAPTER 1 Introduction

the light-sensitive detector, light-intensity variation is translated into “numbers” and

an image array is formed. In the sequel,a series of image processing techniques are

applied leading to line and character segmentation. The pattern recognition soft-

ware then takes over to recognize the characters—that is, to classify each character

in the correct“letter, number, punctuation”class. Storing the recognized document

has a twofold advantage over storing its scanned image. First, further electronic

processing, if needed, is easy via a word processor, and second, it is much more

efﬁcient to store ASCII characters than a document image. Besides the printed

character recognition systems, there is a great deal of interest invested in systems

that recognize handwriting. A typical commercial application of such a system is

in the machine reading of bank checks. The machine must be able to recognize

the amounts in ﬁgures and digits and match them. Furthermore, it could check

whether the payee corresponds to the account to be credited. Even if only half of

the checks are manipulated cor rectly by such a machine, much labor can be saved

from a tedious job. Another application is in automatic mail-sorting machines for

postal code identiﬁcation in post ofﬁces. Online handwriting recognition systems

are another area of great commercial interest. Such systems will accompany pen

computers, with which the entry of data will be done not via the keyboard but by

writing. This complies with today’s tendency to develop machines and computers

with interfaces acquiring human-like skills.

Computer-aided diagnosis is another important application of pattern recogni-

tion, aiming at assisting doctors in making diagnostic decisions. The ﬁnal diagnosis

is, of course, made by the doctor. Computer-assisted diagnosis has been applied to

and is of interest for a variety of medical data,such as X-rays,computed tomographic

images, ultrasound images, electrocardiograms (ECGs), and electroencephalograms

(EEGs). The need for a computer-aided diagnosis stems from the fact that medi-

cal data are often not easily interpretable, and the interpretation can depend very

much on the skill of the doctor. Let us take for example X-ray mammography

for the detection of breast cancer. Although mammography is currently the best

method for detecting breast cancer, 10 to 30% of women who have the disease and

undergo mammography have negative mammograms. In approximately two thirds

of these cases with false results the radiologist failed to detect the cancer, which

was evident retrospectively. This may be due to poor image quality, eye fatigue

of the radiologist, or the subtle nature of the ﬁndings. The percentage of correct

classiﬁcations improves at a second reading by another radiologist. Thus, one can

aim to develop a pattern recognition system in order to assist radiologists with a

“second” opinion. Increasing conﬁdence in the diagnosis based on mammograms

would, in turn, decrease the number of patients with suspected breast cancer who

have to undergo surgical breast biopsy, with its associated complications.

Speech recognition is another area in which a great deal of research and devel-

opment effort has been invested. Speech is the most natural means by which

humans communicate and exchange information. Thus, the goal of building intelli-

gent machines that recognize spoken information has been a long-standing one for

scientists and engineers as well as science ﬁction writers. Potential applications of

such machines are numerous. They can be used, for example,to improve efﬁciency

“03-Ch01-SA272” 17/9/2008 page 3

1.1 Is Pattern Recognition Important? 3

in a manufacturing environment, to control machines in hazardous environments

remotely, and to help handicapped people to control machines by talking to them.

A major effort, which has already had considerable success, is to enter data into

a computer via a microphone. Software, built around a pattern (spoken sounds

in this case) recognition system, recognizes the spoken text and translates it into

ASCII characters,which are shown on the screen and can be stored in the memory.

Entering information by “talking” to a computer is twice as fast as entry by a skilled

typist. Furthermore, this can enhance our ability to communicate with deaf and

dumb people.

Data mining and knowledge discovery in databases is another key application

area of pattern recognition. Data mining is of intense interest in a wide range of

applications such as medicine and biology, market and ﬁnancial analysis, business

management, science exploration, image and music retrieval. Its popularity stems

from the fact that in the age of information and knowledge society there is an ever

increasing demand for retrieving information and turning it into knowledge. More-

over,this information exists in huge amounts of data in various forms including,text,

images, audio and video, stored in different places distributed all over the world.

The traditional way of searching information in databases was the description-based

model where object retrieval was based on keyword description and subsequent

word matching. However, this type of searching presupposes that a manual anno-

tation of the stored information has previously been performed by a human. This

is a very time-consuming job and, although feasible when the size of the stored

information is limited, it is not possible when the amount of the available informa-

tion becomes large. Moreover, the task of manual annotation becomes problematic

when the stored information is widely distributed and shared by a heterogeneous

“mixture”of sites and users. Content-based retrieval systems are becoming more and

more popular where information is sought based on “similarity”between an object,

which is presented into the system, and objects stored in sites all over the world.

In a content-based image retrieval CBIR (system) an image is presented to an input

device (e.g.,scanner). The system returns“similar”images based on a measured“sig-

nature,” which can encode, for example, information related to color, texture and

shape. In a music content-based retrieval system, an example (i.e., an extract from

a music piece), is presented to a microphone input device and the system returns

“similar” music pieces. In this case, similarity is based on certain (automatically)

measured cues that characterize a music piece, such as the music meter, the music

tempo, and the location of certain repeated patterns.

Mining for biomedical and DNA data analysis has enjoyed an explosive growth

since the mid-1990s. All DNA sequences comprise four basic building elements;

the nucleotides: adenine (A), cytosine (C), guanine (G) and thymine (T). Like the

letters in our alphabets and the seven notes in music, these four nucleotides are

combined to form long sequences in a twisted ladder form. Genes consist of,usually,

hundreds of nucleotides arranged in a particular order. Speciﬁc gene-sequence

patterns are related to particular diseases and play an important role in medicine.

To this end, pattern recognition is a key area that offers a wealth of developed tools

for similarity search and comparison between DNA sequences. Such comparisons

“03-Ch01-SA272” 17/9/2008 page 5

1.2 Features, Feature Vectors, and Classiﬁers 5

␮

␴

⫹

FIGURE 1.2

Plot of the mean value versus the standard deviation for a number of different images originating

from class A (



) and class B (⫹). In this case, a straight line separates the two classes.

with a number of patterns, some of which are known to originate from class A and

some from class B.

The ﬁrst step is to identify the measurable quantities that make these two regions

distinct from each other. Figure 1.2 shows a plot of the mean value of the inten-

sity in each region of interest versus the corresponding standard deviation around

this mean. Each point corresponds to a different image from the available database.

It turns out that class A patterns tend to spread in a different area from class B pat-

terns. The straight line seems to be a good candidate for separating the two classes.

Let us now assume that we are given a new image with a region in it and that we

do not know to which class it belongs. It is reasonable to say that we measure the

mean intensity and standard deviation in the region of interest and we plot the cor-

responding point. This is shown by the asterisk (∗) in Figure 1.2. Then it is sensible

to assume that the unknown pattern is more likely to belong to class A than class B.

The preceding artiﬁcial classiﬁcation task has outlined the rationale behind a

large class of pattern recognition problems. The measurements used for the classiﬁ-

cation,the mean value and the standard deviation in this case,are known as features.

In the more general case l features x

, i ⫽ 1, 2, ..., l, are used, and they form the

feature vector

x ⫽ [x

, x

, ..., x

]

where T denotes transposition. Each of the feature vectors identiﬁes uniquely

a single pattern (object). Throughout this book features and feature vectors will

be treated as random variables and vectors, respectively. This is natural, as the

measurements resulting from dif ferent patterns exhibit a random variation. This

is due partly to the measurement noise of the measuring devices and partly to

“03-Ch01-SA272” 17/9/2008 page 6

6 CHAPTER 1 Introduction

the distinct characteristics of each patter n. For example, in X-ray imaging large

variations are expected because of the differences in physiology among individuals.

This is the reason for the scattering of the points in each class shown in Figure 1.1.

The straight line in Figure 1.2 is known as the decision line, and it constitutes

the classiﬁer whose role is to divide the feature space into regions that correspond

to either class A or class B. If a feature vector x, corresponding to an unknown

pattern, falls in the class A region, it is classiﬁed as class A, otherwise as class B.

This does not necessarily mean that the decision is correct. If it is not correct,

a misclassiﬁcation has occurred. In order to draw the straight line in Figure 1.2

we exploited the fact that we knew the labels (class A or B) for each point of

the ﬁgure. The patterns (feature vectors) whose true class is known and which

are used for the design of the classiﬁer are known as training patterns (training

feature vectors).

Having outlined the deﬁnitions and the rationale, let us point out the basic

questions arising in a classiﬁcation task.

■ How are the features generated? In the preceding example, we used the

mean and the standard deviation,because we knew how the images had been

generated. In practice,this is far from obvious. It is problem dependent,and it

concerns the feature generation stage of the design of a classiﬁcation system

that performs a given pattern recognition task.

■ What is the best number l of features to use? This is also a very important

task and it concerns the feature selection stage of the classiﬁcation system.

In practice,a larger than necessary number of feature candidates is generated,

and then the “best” of them is adopted.

■ Having adopted the appropriate, for the speciﬁc task, features, how does one

design the classiﬁer? In the preceding example the straight line was drawn

empirically, just to please the eye. In practice, this cannot be the case, and

the line should be drawn optimally, with respect to an optimality criterion.

Furthermore,problems for which a linear classiﬁer (straight line or hyperplane

in the l-dimensional space) can result in acceptable performance are not the

rule. In general, the surfaces dividing the space in the various class regions

are nonlinear. What type of nonlinearity must one adopt, and what type of

optimizing criterion must be used in order to locate a surface in the right place

in the l-dimensional feature space? These questions concern the classiﬁer

design stage.

■ Finally, once the classiﬁer has been designed, how can one assess the perfor-

mance of the designed classiﬁer? That is,what is the classiﬁcation error rate?

This is the task of the system evaluation stage.

Figure 1.3 shows the various stages followed for the design of a classiﬁcation

system. As is apparent from the feedback arrows, these stages are not independent.

On the contrary,they are interrelated and,depending on the results,one may go back

剩余966页未读，继续阅读

徘徊者GHB

粉丝: 14
资源: 24

《模式识别第四版》：深度探讨有监督与无监督算法

模式识别第四版——学术出版社

模式识别课程讲义-蔡宣平教授主讲

模式识别中的最小最大损失准则

SD2.0标准协议————中文版

商业模式画布——如何找到你的MVP.docx

时间序列分析——基于R（第2版）案例数据

第二版数据仓库工具箱——维度建模的完全指南

时间序列分析——基于R（第2版）习题数据,时间序列分析基于r第二版答案,R language源码.zip

互联网金融给商业银行的影响分析——提交版 - 副本-论文.zip

《模式识别》第四讲：数据聚类的数学定义与应用

最新资源