Handwriting Analysis Using Image Processing And Pattern Recognition

Handwriting Analysis Using Image Processing and Pattern Recognition Naveen Kumar B.M, 2nd Year M.Tech, P.E.S.C.E, Mandya

Abstract – Handwriting analysis is a modern form of psychology that identifies personality traits and human character through handwriting. This paper demonstrates how we can use Image Processing and Pattern Recognition techniques for Handwriting analysis. Even though analysis can be done manually, to get 100% accuracy and to save time I thought of giving a techno-touch to this science. So I came out with this idea of using image processing techniques for handwriting analysis.

1. Introduction

Traditionally, these methods are grouped into two categories: structural methods and feature

In image processing, after acquiring a digitized

space methods. Structural methods are useful in

image, the main tasks are: enhancement or

situations where the different classes of entity

rectification; segmentation; measurement; and

can be distinguished from each other by

data analysis. Image enhancement and image

structural

rectification are often used to emphasize certain

recognition different letters of the alphabet are

features and to remove artifacts respectively.

structurally different from each other. The

Two types of measurements are made: feature

earlier-developed structural methods were the

measurements are taken off individual objects

syntactic methods, based on using formal

which have been defined by a segmentation

grammars to describe the structural are machine

process, and field measurements are obtained

vision methods such as those based on poling

globally from complete images. Finally, these

distribution models, active contours,etc.

information,

e.g.

in

character

feature and field measurements must be analyzed. Pattern recognition by computer is, in general, a complex procedure requiring a variety

2. Image Processing and Pattern Recognition

of techniques that successively transform the iconic data to information directly usable for recognition. Many methods of artificial pattern recognition have been proposed, applicable in general not only to objects in a visual image but also to other types of real world entity.

2.1 Image Processing Sight is a human being’s principle sense. A visual image is rich in information from the outer world and receiving and analyzing such images is part of the routine

activity of human beings throughout their

In image processing, after acquiring a

walking lives. At a more sophisticated level,

digitized

human beings may generate record or transmit

enhancement or rectification; segmentation;

images. These activities together comprise

measurement; and data analysis, as indicated in

image processing.

Figure 2.1. Image enhancement and image

Theories and techniques of image processing originated in the study of optics and

image,

the

main

tasks

are:

rectification are often used to emphasize certain features and to remove artifacts respectively. Two types of measurements are made: feature

optical instruments. However, the advent of

measurements are taken off individual objects

digital computers opened vast new possibilities

which have been defined by a segmentation

for artificial image processing. By the mid-

process, and field measurements are obtained

1960’s, third-generation computers offered the

globally from complete images. Finally, these

speed and storage necessary for practical

feature and field measurements must be

implementation of image-processing algorithms;

analyzed.

and in 1964 the capabilities of digital image processing were spectacularly demonstrated when pictures of the moon transmitted b the Ranger 7 space probe were processed to correct

Enhanceme nt or

segment ation

various types of image distortion inherent in the

Feature measureme nt

on-board television camera. Since that date, the field of image processing

Field measurem ent

has experienced vigorous growth. Digital image processing techniques are used today in a wide range of applications that, although otherwise

Digital Image

unrelated, share a common need for methods

Data analysis

capable of enhancing pictorial information for human

interpretation

and

analysis.

These

applications include: remote sensing; security monitoring;

medical

diagnosis;

automatic

Interpretati on

inspection; radar; sonar; detection of military targets;

robotics;

business

communication;

television enhancement;etc. Fig 2.1 Diagram of Image processing for object In civil engineering, it has been used for structural monitoring, hydrology, and soil microstructure.

recognition

2.2 Pattern Recognition In communication with the outer world, one of

recognition have been proposed, applicable in

the most important goals for human beings is to

general not only to objects in a visual image but

recognize objects. For example, from an image,

also to other types of real world entity.

image set, or image sequence of objects, we need to recognize which directions the objects are oriented toward, where they are located,

Traditionally, these methods are grouped into

how they are arranged, what size and shape they

two categories: structural methods and feature

have, and what sorts of things they are.

space methods. Structural methods are useful in

During the past 30 years, pattern recognition has had a considerable growth. The need for theoretical methods and experimental software and hardware is increasing. Applications of pattern recognition now include: character recognition; target detection; medical diagnosis; biomedical signal and image analysis; remote sensing; identification of human faces and of fingerprints;

reliability

socioeconomics;

archaeology;

analyses; speech

situations where the different classes of entity can be distinguished from each other by structural

information,

e.g.

in

character

recognition different letters of the alphabet are structurally different from each other. The earlier-developed structural methods were the syntactic methods, based on using formal grammars to describe the structural are machine vision methods such as those based on point distribution models, active contours,etc.

recognition and understanding; machine part recognition; automatic inspection; and many In feature-space methods, a set of

others

measurements (typically numerical) is made on each real-world entity (or pattern), and from the Pattern recognition by computer is, in general, a complex procedure requiring a variety of techniques that successively transform the iconic data to information directly usable for recognition. Many methods of artificial pattern

measurement set there is extracted a set of features which together characterize the class of patterns to which the given pattern belongs. The features are regarded as the elements of a vector drawn from the origin in a multi-dimensional

feature space. Ideally, the measurements and

where there exists a continuum of pattern

features are so chosen that (a) the extremities of

classes, rather than a set of discrete classes.

the vectors representing patterns belonging to

Feature-space methods are useful in situations

the same class tend to cluster together in a

where the distinctions between different pattern

region of feature space, and (b) the extremities

classes are readily expressible in terms of

of the vectors representing patterns belonging to

numerical measurements of this kind. Such a

different classes tend to occur in distinct such

situation often exists, for example, in the study

clusters in distinct regions of feature space. A

of soil microstructure, where, for example,

classifier can then assign an unseen real-world

important distinctions between soil particles,

pattern to a particular class according to the

required by soil engineers, are based on such

region of feature space in which the vector

considerations as roundness versus angularity.

representing this pattern falls.

These and other aspects both of the nature of a soil

particle

and

of

soil

structure

lend

themselves to numerical measurement, and there The traditional approach to feature-space pattern recognition is the statistical approach, where the boundaries between the regions representing pattern classes in feature space are found by statistical inference based on a design set

of

sample

patterns

of

known

was an urgent need for numerically based classification for immediate comparison with numerical properties of the soil. The feature space approach is the one that has therefore been used in this research.

class

membership. An unseen pattern can then be classified simply by determining the region of feature space in which it lies. An alternative approach is to use a mathematical or physical

3.

Design

Issues

in

Pattern

Recognition System

model of the pattern generating mechanism to

Unclassified specimens are the specimens

predict the regions: this approach is useful in

which are to be classified. Pattern analysis is the

situations where it is costly or impossible to

process of extracting the characteristics of the

obtain sufficient numbers of design samples to

specimens;

allow statistical conclusions to be drawn form

measurements

them with any degree of confidence. A third

Training is specimens whose class membership

possibility, which appears to be due to the

is taken as known a priori; in almost all cases, it

author, is to choose features sot hat the total

is the set of characteristics obtained from these

hyper volume of feature space within which

specimens which is used.

these characteristics or

structured

might be

observations.

feature points can occur is know a priori. The whole of feature space can then be partitioned

A priori definitions are definitions of the

according to some suitable scheme for the

classes which have been set up in advance,

problem in hand. This approach might be useful

either on the basis of some theoretical analysis

or in an entirely arbitrary fashion depending on

A pattern is the set of characteristics which is

the nature of the problem. The criteria are

inherent in a sample. These patterns may be

definitions of the closeness with which an

taken from real samples; BT some synthetic

unclassified specimen must match the definition

patterns designed to test the system may be

of a particular class in order to be placed in that

included. Here, pattern analysis is the process of

class; if no class is sufficiently closely matched,

extracting the actual set of characteristics to be

the specimen ma be rejected i.e. not placed in

used in the classification.

any class. These criteria may be set to broad or narrow limits depending on the use to which the results of the classification will be put.

The a priori definitions, training samples, and criteria, are the current versions of these parts of the system; but during the design

Decision making is the process of

process, these may not yet have been finalized.

comparing the actual characteristics with those

The results are then inspected to see whether

on which the classification is to be based. In

they are judged to be satisfactory. If not, the

some cases, it is appropriate to monitor the lack

error is fed back to modify the current versions

of fit, i.e. the error, and to use this to modify the

of the parts of the system. Figures 3.1(a) and

set of characteristics which is actually being

3.2(b) is a simplified and generalized view of

used for classification.

the process of designing a pattern recognition system.

Classifier (Analysis Phase) Error

classes

Decision

Unclassified Pattern

Analysis

making

Specimens

Criteria

A priori definitio

Training samples

Figure 3.1(a) Operation of a pattern recognition system

Patterns

Pattern recognitio

A priori definition s

Error

Training samples

Decision making

n

classes

Criteria

Figure 3.1(b) Designing a pattern recognition system.

4. Example System: Handwriting Analysis

personality is a subcategory of general psychology. Personality is a cluster of character traits and the

Handwriting analysis is a modern form of psychology

corresponding behaviors based on those character

that identifies personality traits and human character

traits. Handwriting analysis is the quickest way to

through handwriting. Personality is the term for behavior and character. In the study of psychology,

accurately discover someone’s true personality out of

To get accuracy we will be comparing all the letters,

any formal psychological test on the market today.

instead of stopping at one or two letters.

Handwriting analysis is categorized into to groups,

If the letters are not segregated it is not possible to

those are

apply this method.

1. Micro Analysis

2. Macro Analysis

We can reveal more than 100 personality traits using

In this we need find the size of the writing as well as

this analysis. Even though not all the letters

slant of the writing. This can be done by using

contributes to this analysis, we will need them at

Emotional Gauge.

some point of time. AB

With the help of image processing and pattern

B C

CD

DE

recognition techniques we would be able to segregate each letters from the handwriting samples.

E+

After that we would be comparing the captured image (i.e. letter) with the pre-defined set of images. If that particular criterion matches then we will say

Large Averag e Small

that the man has that particular trait. So, we need to maintain a database for storing the pre-defined images. Before we compare the captured letter with the pre-defined image, we will place a control point (collection of pixels) on the captured letter, to avoid the conflict between two over lapped letters

Fig 4.1 Emotional Gauge. The first step is to decide which letters you are going to use to measure slant. Some letters have three or more strokes to constitute the entire letter. Measuring for slant is actually measuring one stroke; so many letters could have more than one measurable stroke,

Let us see with an example, consider the letter ‘t’

or none. The easiest letters to measure are the cursive

which is captured from the handwriting sample.

t, b, l, m, n, r, s, d, h, and k.

5. Control points

Applications

and

Limitations

of

Handwriting Analysis 5.1 Applications of Handwriting Analysis

Here the control points are considered as edges for

There are many uses of handwriting analysis. Below

that letter, we would be only concentrating only up to

are a few of the most popular applications we use

those edges not beyond.

today. You will find more. •

Dating and Socializing

•

Employee hiring and human resources

•

•

Police profiling

•

Self improvement and professional speaking

•

Counselor,

or captured through digital cameras.

therapist,

and

coaching

applications.

6. Conclusion The work presented here is mainly aimed at analyzing the people’s handwriting to know more

5.2 Limitations of Handwriting analysis

about them through the computer by making use of

Below are a few of the limitations of Handwriting analysis •

Handwriting samples must be scanned

the science called graphology. I hope the tool which I am designing will reach out

Handwriting does not reveal the AGE of the

the people for the best utilization. I am aiming at the design of the tool which will be very much user

writer

friendly rather than a messy one. •

Handwriting does not reveal the Gender of the writer

•

samples provided for analysis are legible to read and

Handwriting does not reveal if the writer has written using left or the right hand or any other part of the body.

•

cannot

be

found

in

the

handwriting •

not a printed one as well. The future work encompasses finding out the hell traits and giving the remedy for it.

Caste, religion, race, creed or religious preferences

The proposed scheme assumes that the handwriting

References [1] Daisheng Luo, Pattern Recognition and Image

The future cannot be predicted with the handwriting.

Processing, Horwood Publishing 1998. [2] Jorge Boliva, Nearest Neighbor in Pattern Recognition, IEEE/ACM, Jan 1990.

Limitations when we do analysis using image processing techniques are: •

We cannot know the pressure of

Dealing difficult.

international conference on Pattern

[4] Bart A Baggett, Handwriting Analysis & Success Secrets

with

Recognition,

September 1992

writing. •

[3] Laveen Kanal, On Patterns and Categories, 11th

illegible

writing

is