Pattern Recognition #1 - Gulraj

MuhammadGulraj
BS Computersystemengineering, GIKI,Pakistan
MS Computersystemengineering, UETPeshawar,Pakistan
1
Pattern Recognition
Name: Muhammad Gulraj
Muhammad.gulraj@yahoo.com

MuhammadGulraj
2
Introduction
In real world human beings identify certain things using pattern recognition. It is a built in ability
in human beings, e.g. humans recognize different melodies, faces, words and images using
their innate abilities of pattern recognition.
In computer science Pattern Recognition ‘is taking a decision or inference on the basis of
some input data using the patterns of data’, e.g. an Email system decides on the basis of certain
patterns, whether a specific email is spam or non-spam. Similarly using the patterns of DNA it
can be inferred whether a patient have chances of breast cancer or not.
Input Data Results
Pattern recognition assigns label to input values using classification or regression, which are the
main forms of Pattern recognition. Classification is usually used for discrete class labels in
which certain decisions are made by assigning a label to each input value from the given
classes. Supervised learning is a type of classification because the training set has been
provided. Deciding whether an email is spam or non-spam is an example of
Classification/Supervised learning.
Pattern recognition
system

MuhammadGulraj
3
The above figure shows an example of Classification. The figure shows that a patient having
small tumor size and young age have less chances of breast cancer (blue color), than the one
having large tumor size and older age (red color). Regression is used for labeling continuous
values.
In Unsupervised learning there is no training set/data provided on which the system can be
trained. Search engines usually implements unsupervised learning algorithms for
indexing/clustering of information. Pattern recognition tries to find the best answer using
statistical and probabilistic variations.
Applications
Pattern recognition has been used in security systems, medical diagnosis, search engines, data
mining and speech recognition systems, optical and hand written character recognition systems,
identification systems, robotics and analysis of astronomical data.
Pattern recognition has been vastly used in Security systems. These systems are normally
used in banks, offices, military and government installations. Some security systems which use
pattern recognition are as follows.
 Iris recognition system uses the patterns of iris in human eyes to recognize
individuals. Every human being have unique iris pattern inside his eye on the basis of
which a human being can be recognized.
 Face recognition system is another system that uses the patterns of skin and other
features to identify humans in images and videos.

MuhammadGulraj
4
 Another system that is used for security purposes is Finger print recognition system
which uses three different patterns namely whorl, arch and loop of thumb to identify a
person.
Pattern recognition is also used in medical systems to diagnose patients. Systems such as
MRI scanning and computer aided diagnosis systems CAD are using pattern recognition
techniques to identify different diseases.
Search engines e.g. yahoo, Google, Ask and Bing also use pattern recognition techniques to
identify and index information for users.
Data mining is another field in which pattern recognition has been used. Data mining is
basically used for extracting useful information from huge data sets.
Pattern recognition is also used in speech recognition and speech tagging. Speech
recognition system is used for security purposes as well as research in natural language
processing.
Robotics is using pattern recognition to inculcate supervised as well as unsupervised learning
in robots. Scientists are working on super intelligent robots, which can understand the
environment and act accordingly. NASA has developed pattern recognition robots which can be
used for navigating and studying environment of other planets such as Mars. Autonomous and
Semi-autonomous cars have also been introduced using pattern recognition techniques.
Industrial automation uses pattern recognition techniques to identify items that have some
fault from the rest. These techniques have reduced work hours and enhanced efficiency.
Astronomers and scientists use pattern recognition techniques to identify and study different
galaxies. Usually techniques such as decision tree are used for star – galaxy classification.

MuhammadGulraj
5
Apart from the above mentioned systems, pattern recognition techniques and algorithms are
used in image processing, computer vision, machine learning, artificial intelligence, cognitive
science and psychology.
Basic steps of pattern recognition task
The basic steps that involves in pattern recognition task are
 Data acquisition using sensors.
 Pre- processing such as segmentation and contrast adjustment.
 Feature extraction from the Pre-processed data.
 Classification on the basis of features extracted.
 Post processing which includes cost to improve system’s performance.
 Decision on the basis of the above steps
Data
Acquisition
Pre -
Processing
Feature
Extraction
Classification
Post
Processing
SENSOR
DecisionInput Data

MuhammadGulraj
6
Techniques
The system usually acquires data using different sensors such as camera, microphone or
scanner. These sensors normally convert analog or continuous data into discrete form so that it
can be used in computer. The data acquired using sensors can be used as input data as well as
a training set for the system.
The second step is Pre-processing. Pre-processing is a process in which the system perform
some operations on the input data to make it useful for the next step. There are several
techniques to make this data useful, e.g. if the input data that is acquired is an image or video
then this image can be divided into different segments or pixels (Segmentation). This image can
also be made useful by adjusting the contrast of the image.
After pre-processing the data is passed to the next step which is called Feature Extraction.
Feature extraction is a domain specific problem and different methods can be applied to extract
features, e.g. in a system that is used for Human recognition, we can extract features such as
shape, hair, legs, skin color, nose, eyes and hands using edge detection, motion detection and
color intensities. Contrast equalization and Gabor filter can be used to extract features from an
image in face recognition system.
The Classification uses the features extracted by the system to assign classes to each point.
The system will have to establish some threshold on the basis of which a certain class can be
assign to every data point. A system can use one or more classifiers depending upon the
problem. There are different classifiers available which can be used such as Bayes classifier,
linear classifiers, nearest neighbor classifier and support vector machines. Bayes classifier is
the most common classifier which is used in classification.
Post processing is usually used for improvement in performance using different techniques
such as minimizing the cost of classification. Post processing can enhance the quality of
decision. Using the above steps the system decides to perform a particular action.

MuhammadGulraj
7
Areas and application for research
Pattern recognition is a vast and interesting field of study which is interrelated with Artificial
intelligence and machine learning. There are enormous opportunities of research in this field.
Research area includes feature extraction, classification, and discriminant analysis, analysis of
astronomical data using clusters, human identification, image/video analysis, data mining and
business intelligence, speech recognition, optical character recognition OCR, medical diagnosis
systems, industrial automation, autonomous vehicles, robotics and error/cost estimation.
Robotics is another field which uses pattern recognition extensively. Rover 1 which has been
recently sent to Mars by NASA to identify and navigate the environment uses Pattern
recognition. Robots can also be used for fire-fighting. Advance research is going on in robotics
as well.
Medical and cognitive science is another field in which application development and research
has been done using pattern recognition. In cognitive science patterns of human behavior and
their actions according to these patterns are studied. Psychologists use these patterns to
identify human behavioral disorders. Medical diagnosis decision support system is another field
of medical/pattern recognition in which research is carried out.
Pattern recognition is an important subject of research in security systems. There are many
security systems which uses pattern recognition techniques. But these systems have also some
hacks, so scientists are constantly trying to develop new algorithms and techniques to find new
and enhanced solutions.
Unlimited recognition such as cursive script and continuous speech/image recognition is still a
distant dream for scientists and a lot of work needs to be done to achieve that dream.

MuhammadGulraj
8
References
1. Andrew Ng (2013), an online course for Machine learning, Stanford University,
Stanford, https://class.coursera.org/ml-004/class.
2. Duda and Hart, Pattern Classification (2001-2002), Wiley, New York.
3. Ying Cui and Zhong Jin, Facial feature points (2012),
http://www.jprr.org/index.php/jprr
4. Ioannis Dimou and Michalis Zervakis, On the analogy of classifier ensembles,
http://www.jprr.org/index.php/jprr
5. Nair, H., A system for pattern recognition and pattern summarization in multiband
satellite images, http://www.springer.com/computer/image+processing/journal/11493

Pattern Recognition #1 - Gulraj

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Pattern Recognition #1 - Gulraj (20)

Pattern Recognition #1 - Gulraj