Académique Documents
Professionnel Documents
Culture Documents
ABSTRACT: Object Classification is an important task within the field of computer vision. Image classification refers to the
labelling of images into one of a number of predefined categories. Classification includes image sensors, image pre-processing,
object detection, object segmentation, feature extraction and object classification.Many classification techniques have been
developed for image classification. In this survey various classification techniques are considered; Artificial Neural
Network(ANN), Decision Tree(DT), Support Vector Machine(SVM) and Fuzzy Classification
Keywords: Image Classification, Artificial Neural Network, Decision Tree, Support Vector Machine, Fuzzy Classifier.
C. Training: Selection of the particular attribute
which best describes the pattern.
I. INTRODUCTION D. Classification of the object-Object
Classification between the objects is easy task for classification step categorizes detected objects
humans but it has proved to be a complex problem for into predefined classes by using suitable method
machines. The raise of high-capacity computers, the that compares the image patterns with the target
availability of high quality and low-priced video patterns.
cameras, and the increasing need for automatic video
II. IMAGE CLASSIFICATION APPROACHES
analysis has generated an interest in object classification
algorithms. A simple classification system consists of a
camera fixed high above the interested zone, where Various image classification approaches are defined
images are captured and consequently processed. briefly:
Classification includes image sensors, image pre-
DIGITAL DATA
processing, object detection, object segmentation, feature
extraction and object classification. Classification system PREPROCESSING
consists of database that contains predefined patterns that
FEATURE
compares with detected object to classify in to proper EXTRACTION
category. Image classification is an important and
challenging task in various application domains, SELECTION OF
TRAINING DATA
including biomedical imaging, biometry, video-
surveillance, vehicle navigation, industrial visual DECISION AND
inspection, robot navigation, and remote sensing. CLASSIFICATION
Classification process consists of following steps:
A. Pre-processing- atmospheric correction, noise CLASSIFICATION
OUTPUT
removal, image transformation, main
component analysis etc. POST-PROCESSING
B. Detection and extraction of a object-
Detection includes detection of position and ACCURACY
ASSESSMENT
other characteristics of moving object image Fig 1 Steps for image classification
obtained from camera. And in extraction, from
the detected object estimating the trajectory of 1) On The Basis Of Characteristic Used:
the object in the image plane. A. Shape-based:
Copyright to IJARCCE www.ijarcce.com 1005
ISSN (Print) : 2319-5940
ISSN (Online) : 2278-1021
This methods make use of the objects’ 2D spatial The spectral value of each pixel is assumed to be a
information. Common features used in shape-based linear or non-linear combination of defined pure
classification schemes are the points (centroid, set of materials called end members, providing
points),primitive geometric shapes(rectangle or ellipse), proportional membership of each pixel to each end
skeleton, silhouette and contour. member. Subpixel classifier has the capability to
B. Motion-based: handle the mixed pixel problem, suitable for
This methods use temporal tracked features of objects for medium and coarse spatial resolution images.
the classification. Example: spectral mixture analysis, subpixel
classifier, Fuzzy-set classifiers.
2) On The Basis Of Training Sample Used: C. Per-field classifier:
A. Supervised Classification: The per-field classifier is intended to handle the
The process of using samples of known informational problem of environmental heterogeneity, and also
classes (training sets) to classify pixels of unknown improves the classification accuracy. Generally used
identity. Example: minimum distance to means by GIS-based classification approaches.
algorithm, parallelepiped algorithm, maximum likelihood D. Object-oriented classifiers:
algorithm Pixels of the image are united into objects and then
B. Unsupervised Classification: classification is performed on the basis of objects.It
In this type of classification is a method which examines involves 2 stages: image segmentation and image
a large number of unknown pixels and divides it into classification Image segmentation unites pixels into
number of classes based on natural groupings present in objects, and a classification is then implemented on
the image values. Computer determines spectrally the basis of objects. Example: e Cognition.
separable class and then defines their information value.
No extensive prior knowledge is required. Example: K- 5) On The Basis Of Number Of Outputs For Each
means clustering algorithm. Spatial Element:
A. Hard Classification:
3) On The Basis Of Assumption Of Parameter on Also known as crisp classification In this each pixel
Data: is required or forced to show membership to a
A. Parametric classifier: single class.eg maximum likelihood, minimum
The parameters like mean vector and covariance matrix distance, artificial neural network, decision tree,
are used.There is an assumption of Gaussian distribution. and support vector machine.
The parameters like mean vector and covariance matrix B. Soft classification :
are frequently generated from training samples.Example: also known as fuzzy classification In this each pixel
Maximum likelihood, linear discriminant analysis. may exhibit numerous and partial class
B. Non Parametric classifier: membership. Produces more accurate result.
There is no assumption about the data. Non-parametric
classifiers do not make use of statistical parameters to 6) On The Basis Of Spatial Information
calculate class separation. Example: Artificial neural A. Spectral Classifiers:
network, support vector machine, decision tree classifier, This image classification uses pure spectral
expert system. information .Example: Maximum likelihood,
minimum distance, artificial neural network.
4) On The Basis Of Pixel Information Used: B. Contextual Classifiers
A. Per pixel classifier: This image classification uses the spatially
Conventional classifier generates a signature by using neighbouring pixel information. Example:
the combination of the spectra of all training-set frequency-based contextual classifier.
pixels from a given feature. the contributions of all C. Spectral-contextual classifiers:
materials present in the training-set pixels is present This classification uses both spectral and spatial
in the resulting signature. It can be parametric or non- information initial classification images are
parametric the accuracy may not meet up because of generated using parametric or non-parametric
the impact of the mixed pixel problem. Example: classifiers and then contextual classifiers are
maximum likelihood, ANN, support vector machine implemented in the classified images. Example:
and minimum distance. combination of parametric or non-parametric and
B. Subpixel classifiers: contextual algorithms.
Fuzzy Measure In Fuzzy It uses Stochastic design and which classes are
classification, approach. training. clearly
various stochastic Performance and Provides distinguishable.
associations are accuracy depends hierarchical Becomes complex
determined to upon the threshold associations calculation when
describe selection and fuzzy between input various values are
characteristics of integral. variables to undecided and/or
an image. The forecast class when various
various types of membership and outcomes are
stochastic are provides a set of correlated.
combined (set of rules n are easy to
properties) in interpret.
which the Simple and
members of this computational
set of properties efficiency is good.
are fuzzy in nature. Support It gains flexibility Result transparency is
It provides the Vector in the choice of low.
opportunity to Machine the form of the Training is time
describe different threshold. consuming.
categories of Contains a non- Structure of algorithm
stochastic linear is difficult to
characteristics in transformation. understand
the similar form. It provides a good Determination of
generalization optimal parameters is
capability. not easy when there is
The problem of nonlinearly separable
over fitting is training data.
TABLE II eliminated.
Advantages and Disadvantages of Different Classification Reduction in
Techniques computational
complexity.
Simple to manage
Classification Advantages Disadvantages decision rule
method complexity and
Artificial It is a non-parametric It is Error frequency.
Neural classifier. semantically Fuzzy Efficiently handles Without priori
network It is an universal poor. Measure uncertainty. knowledge output is
functional The training of properties are not good
approximator with ANN is time describe by precise solutions
arbitrary accuracy. taking. identifying various depends upon
capable to present Problem of over stochastic direction of decision.
functions such as OR, fitting. relationships.
AND, NOT Difficult in
It is a data driven self- choosing the
adaptive technique type network
efficiently handles architecture. IV. CONCLUSION
noisy inputs This paper attempts to study and provides a brief
Computation rate is
knowledge about the different image classification
high
approaches and different classification methods. Most
zz common approaches for image classification can be
categories as supervised and unsupervised, or parametric
and nonparametric or object-oriented, subpixel, per-pixel
and perfield or spectral classifiers, contextual classifiers
Decision tree Can handle non- The usage of
and spectral-contextual classifiers or hard and soft
parametric hyperplane decision
training data boundaries parallel to classification. This survey gives theoretical knowledge
Does not required the feature axes may about different classification methods and provides the
an extensive restrict their use in
Copyright to IJARCCE www.ijarcce.com 1008
ISSN (Print) : 2319-5940
ISSN (Online) : 2278-1021