fingerprint image classification and retrieval using ......fingerprint image classification and...
TRANSCRIPT
Fingerprint Image Classification and Retrieval using
Statistical Methods
Ph.D. Synopsis
Submitted To
Gujarat Technological University
For the Degree
of
Doctor of Philosophy
in
Computer IT Engineering
By
Sudhir P Vegad
Enrollment No: 119997107012
Supervisor:
Dr. Tanmay Pawar
Professor & Head
Electronics Department
BVM Engineering College
Vallabh Vidyanagar
Foreign Co-Supervisor:
Dr. Arun Somani
Associate Dean for Research
IOWA State University
Ames, IA
Table of Contents
Page
1. Title of the Thesis and Abstract............................................................................. 1
2. Brief description on the state of the art of the research topic................................ 2
3. Objective and Scope of the work........................................................................... 8
4. Original contribution by the thesis......................................................................... 9
5. Methodology of Research, Results / Comparisons................................................ 10
6. Achievements with respect to objectives............................................................... 13
7. Conclusions............................................................................................................ 13
8. List of publication arising from the thesis............................................................. 15
9. References.............................................................................................................. 16
1 | P a g e
1. Title of the Thesis and Abstract
1.1 Title of the Thesis
Fingerprint Image Classification and Retrieval using Statistical Methods.
1.2 Abstract
Human fingerprints are popular biometrics due to its characteristics like universality,
uniqueness, permanence, collectability and acceptability. Fingerprints are rich in details
which are known as minutiae, which can be used as identification marks to uniquely identify
the fingerprint for various purposes. There are four basic patterns of fingerprint ridges; arch,
left loop, right loop and whorl.
Fingerprint classification is an important indexing scheme to narrow down the search of
fingerprint database for efficient large-scale retrieval. It is still a challenging problem due to
the intrinsic class ambiguity and the difficulty for poor quality fingerprints. Fingerprint
retrieval is a pre-requisite step for many applications like recognition and registration.
Feature extraction plays a very important role in retrieval process. The work presented in this
research has two phases; classification and retrieval. In classification phase, scale and rotation
invariant features are extracted to enhance the local information from the fingerprint images
for powerful representation. These features strongly represent the similarities of fingerprint
images of the similar class. Classification is carried out by applying these features in
classification models. The retrieval phase match the fingerprint image features using distance
matching methods to retrieve similar images from the class declared from classification
phase. The experimental results and comparisons on FVC2000, FVC2002, FVC 2004 and
NIST-4 databases have shown the effectiveness and superiority of the proposed method for
fingerprint classification and retrieval.
2 | P a g e
2. Brief description on the state of the art of the research topic
The identification of a person with an AFIS (automated fingerprint identification system)
requires a comparison of his/her fingerprint with all the fingerprints in a database [1]. This
database may be very large in many forensic and civilian applications, which leads to long
processing time and deteriorated accuracy; hence it is unsuitable in real time applications. A
common strategy to speed up the search is to divide the fingerprint database in to a number of
groups that have similar properties. Fingerprint classification that assigns a fingerprint to a
class (based on predefined classes) is an effective way.
Fingerprint classification has generated great interest due to its importance and intrinsic
difficulty and there are number of approaches proposed in the literature. A typical fingerprint
classification algorithm usually extracts a representative feature set to capture the
individuality of each fingerprint and then does some strategies to determine the fingerprint
class.
2.1 Fingerprint Classification
The first phase of is of fingerprint image classification. Most of the classification algorithms
are classifying fingerprint image in five classes; Arch, Left Loop, Right Loop and Whorl. Kai
Cao, Liaojun Pang, Jimin Liang amd Jie Tian [1] presented a hierarchical classifier which
estimates the orientation of fingerprint using root filtering. Diffusion model is used to
regularize the orientation to make classification rotation invariant. The classification is
performed in five stages based on the singular point detection. In the first stage, arch is
distinguished using complex filter responses. The second stage distinguishes whorl by using
core point and ridge line flow classifier. The third stage identifies arch and whorl by k-NN
classifier by using orientation fields and complex responses. Ridge line flow classifies loop
arch. SVM is used for final classification.
Nannevan Noord and EricPostma [4] proposed the use of conventional neural networks to
make image classification scale-invariant. The combination of scale-variant and scale-
invariant features are used to attempt the image classification using task-relevant
characteristics at multiple scales. Significant improvement in image recognition is stated due
to the classification phase prior to recognition.
A fingerprint classification algorithm which uses Adaboost learning method is proposed by
Manhua Liu [5]. Singularities are detected using complex filters responses at multiple scales
3 | P a g e
and feature vector is constructed at each scale with relative position and direction of
singularities. Adaboost learning method is applied on the decision tree to generate classifier
for fingerprint classification. The experiments are carried out on NIST-4 database.
Shing Chyi Chua, Eng Kiong Wong and Alan Wee Chiat Tan proposed an algorithm for
fingerprint classification by detecting the singular point via quantization [18]. A singular
point is detected using a quantization approach on the orientation field of fingerprint image
and core-delta points are located by the changes of the gray level around a 2 x 2 window. The
unwanted singular points are removed with the application of edge-trace-cum-core-delta-
pairing algorithm and merging-and-pruning algorithm. The experiments have been carried
out on NIST-4 database for 5-class and 4-class classification.
Crease features are used for classification by observing crease features as strips in the
fingerprint images [22]. Masking is used to remove noise from the extracted crease features.
Minimum bounding box and convex hull is used to determine the size of every crease as it
this feature is found robust to distortion and classification. The experiments are performed on
Hong Kong PolyU High Resolution Fingerprint.
An approach is discussed by S. Govindaraju and G. P. Ramesh Kumar for retrieving medical
images using robust features [3]. The method applies the SURF algorithm in the detection,
description, extracting reference images and matching feature points in the image
respectively. In the process of feature point matching, the false matching points are
eliminated. Clusters are formed with the descriptors by applying BoF, centroids are formed
and histogram is generated as an pre-requisite step for retrieval.
Local Binary Pattern (LBP) features are used by Guoying Zhao, Timo Ahonen, Jiri Mata, and
Matti Pietikainen [14] for rotation invariant description of image and video description. The
LBP operator is extended to use the circular neighborhood of different sizes. To make
rotation invariant representation, bit pattern is circularly rotated to the minimum value. The
experiments are carried out on Outex_TC_00012 database.
The magnitude component of LBP is utilized in leaf image classification by Anilkumar
Muthevi [19] in which evaluate a value before generating bit pattern, which find the
relativeness within a 3x3 neighborhood by using a value called threshold that is multiplied
with the corresponding weights. The experiments are carried out on the various leaves
databases to determine texture features using LBP approach.
4 | P a g e
Image classification using Gray Level Co-occurrence Matrix (GLCM) is proposed by A
Suresh and K L Shanmuganathan [10]. GLCM is calculated from original texture image and
the differences are calculated along the first singleton dimension of input texture image.
Contrast, homogeneity, energy and correlation are calculated from GLCM. k-NN classifier is
used for classification. System performance is evaluated by using Brodatz database and
compared with the methods PSWT, TSWT, the Gabor transform and Linear Regression
Model.
Another approach of texture extraction from image is proposed by P. Mohanaiah, P.
Sathyanarayana and L. GuruKumar [9]. Features namely, Angular Second Moment,
Correlation, Inverse Difference Moment and Entropy are computed. The experimental results
are shown stating the texture features having high discrimination accuracy and requires less
computation time. Hence GLCM is claimed having reduced image compression time in the
process of converting RGB image to gray level image.
Ray-I Chang, Shu-Yu Lin, Jan-Ming Ho, Chi-Wen Fann and Yu-Chun Wang presented k-
means and k-NN clustering algorithms for content based image retrieval system. Training of
the system is performed by applying segmentation and grid generation, followed by k-means
clustering and neighborhood table generation. The query image is going the same stages
except k-NN clustering rather than k-means clustering. Future enhancements are also
proposed for improved grid module to minimize the loss of information due to quantization.
Tran Son Hai, Computer Science Department and Nguyen Thanh Thuy have proposed the
classifier k-NN for the classification of facial expressions [36]. Independent Component
Analysis (ICA) is used to extract facial features. k-NN classifier uses the ratio features for
classification along with Artificial Neural Networks. Basic seven universal facial expressions
are the classes used for classification.
2.2 Fingerprint Retrieval
Fingerprint Retrieval is process of finding similarities among the query fingerprint image and
that of from the fingerprint database. Using strong feature extraction methods which can
represent the fingerprint images uniquely are used before matching. Distance matching
strategies are used to find the similarities among the fingerprint images and n closest images
are presented as a result [15][28].
5 | P a g e
K. C. Leung and C. H. Leung [11] have proposed the method for improving fingerprint
retrieval by artificially expanding the set of training samples of fingerprints using spatial
modelling technique. Bays classification approach is used to classification and recognition of
fingerprint images. The experiments have been carried out on FVC and NIST-4 databases.
Yashika Birdi and Er. Jagroop Kaur proposed a method based on wavelets in which first of
all an input fingerprint image is decomposed upto three levels using DWT and then textural
features are extracted [15]. Mean, standard deviation and energy computed to build feature
vector. Similarity of fingerprint images is derived using Chi-square approach. Evaluation of
the system is done through various parameters like Specification, Sensitivity, Positive
Predictive Value, Negative Predictive Value, Recognition Rate and Accuracy.
Another wavelet based retrieval is discussed by Javier A. Montoya Zegarra, Neucimar J.
Leite and Ricardo da Silva Torres [16]. Similar to the above approach, the fingerprint image
is decomposed in different levels in wavelet domain and feature vector is built using mean
and standard deviation. Different wavelets used in their study include: Gabor wavelets, tree-
structured wavelet decomposition using both orthogonal and bi-orthogonal filter banks, as
well as the steerable wavelets. Square Chord Similarity measure is used for retrieval.
Among the features used for accurate retrieval, SURF is popular due to its fast feature
extraction approach. Sukhmanjeet Kaur and Mr. Prince Verma proposed and integration of
Artificial Neural Network (ANN) using SURF and SVM [28]. The approach applies SURF
for the detection description of the features. During the feature matching phase, false
matching points are eliminated. The features built with SURF is applied to SVM and ANN
for further classification.
SURF has been presented as a novel scale and rotational invariant detector and descriptor by
Herbert Bay, Andreas Ess, Tinne Tuytelaars, Luc Van Goola [33]. For image convolution
stages, the image is converted in Integral Image which makes the computations for the
summation of any rectangular box faster than any other feature extraction approaches. SURF
uses Hessian matrix-based measures for the detectors and detectors. SURF is applied for the
camera calibration for image registration and object recognition.
Nursabillilah Mohd Ali, Soon Wei Jun, Mohd Safirin Karis, Mariam Md Ghazaly and Mohd
Shahrieel Mohd Aras have presented an approach of classification of the object by extracting
6 | P a g e
SURF features and statistical classifier [37]. The recognition of the object is performed by
extracting the SIFT features.
E.N. Mortensen, Hongli Deng and L. Shapiro proposed SIFT as a descriptor which can find
matches between features with unique neighborhood along with the global context vector for
detecting the matching points of multiple images of a scene [35]. The SIFT helps in the
detailing of local descriptors which is enhance by augmenting with global context vector to
minimize the mismatches when multiple local descriptors are similar. Experiments are
performed on veriety of scene images and compared matching accuracy between the SIFT
with global context to that without.
Hiroharu Kato and Tatsuya Harada have presented the Bag of Features (BoF) to reconstruct
images. BoF is used as a histogram histogram of quantized descriptors extracted densely on a
regular grid at a single scale [34]. Large-scale image database is used to estimate the spatial
arrangement of local descriptors. They proposed a heuristic method to optimize jigsaw puzzle
problem with adjacency and global location costs. BoF are extracted from various image
categories and demonstrated the reconstruction of original images.
Chih-Fong Tsai presented the use of BoF for image annotation. The proposed approach is to
help users to search the images through keywords but eliminates the labeling of images
manually [32]. As image annotation problem can be regarded as an image classification
problem, BoF classifier is used to represent global features of images. The detection of region
of interest is performed and local descriptors are computed. The descriptors are quantized to
words to generate vocabulary. BoW features are constructed by finding the occurrence of
each specific word in the image.
3. Objective and Scope of work
A fingerprint classification system must be able to analyze and recognize the class, i.e. Arch,
Left loop, Right loop and Whorl of a fingerprint image. The system should also be able to
recognize fingerprint images accurately without human intervention. One of the objectives of
this work is to develop a method which recognizes the fingerprint class using the statistical
features extracted from the fingerprint images. Another objective is to develop a method of
uniquely representing the fingerprint image so that most relevant images are retrieved in the
retrieval phase which contains the similar fingerprint image.
The major objectives of this research work are summarized as below:
7 | P a g e
To develop an algorithm for fingerprint features extraction to recognize fingerprint
class. Scale and rotation invariant features are extracted and given to the classifiers for
class identification. The algorithm is to be tested on NIST – 4 standard database
which contains 8-bit gray scale fingerprint image pairs.
To develop an algorithm which extract robust features such that it speeds up the
process of feature extraction for retrieval. The algorithm should represent the local
interest points which makes reduction in the size of feature vector and still strongly
represent the class of fingerprints uniquely. The algorithm is to be tested on NIST – 4
database.
To develop an algorithm which uses hybrid feature extraction technique to extract
features such that the given fingerprint image can be matched with similar image class
images with improved accuracy.
The scope of the work:
The work is tested on two standard databases NIST – 4 and FVC. NIST – 4 contains
image pairs while FVC databases contains eight images of fingerprints.
The work is taking in account the acquisition methods of fingerprint images. The
training and testing of the images is performed with the same databases.
4. Original contribution by the thesis
In this thesis, method of fingerprint image classification and retrieval system is suggested
with variations and combinations of feature extraction, classification and retrieval techniques.
Images from two different databases are used. One is NIST special database 4 (NIST-4),
Fingerprint Verification Competition databases (FVC2000, FVC2002 and FVC2004) and the
fingerprint images captured using optical sensor with the help of volunteers.
The main contributions of this thesis are summarized as follows:
The performance of the classification method strongly depends on the accuracy of the
features extracted. A hybrid feature extraction method is developed that uses the
combinations of different feature extraction algorithms. The feature extraction
algorithms using LBP and GLCM are used collectively to extract features from
fingerprint images. The concatenated features are given to the Support Vector
8 | P a g e
Machine classifier for classification. K-nearest neighbor (k-NN) classifier is also used
for the comparison of the classification results.
BoF classifier which is popular for document classification is experimented with the
fingerprint images to check the possible improvements in the accuracy level of
classification. The Speed-Up Robust Feature (SURF) algorithm is used to extract the
features to test BoF classifier along with the hybrid feature extraction method. Both
the classification algorithms are evaluated on NIST-4 and captured fingerprint images
databases.
It is essential to extract the features which are scale and rotation invariant to retrieve
fingerprint images with accuracy. In retrieval phase, Scale Invariant Feature
Transform (SIFT) features and SURF are used collectively to extract features from the
fingerprint images of the same class. The concatenated features are extracted from
these algorithms are given to BoF and other distance matching algorithms for
indexing. The retrieval algorithms are evaluated on NIST-4, FVC and captured
fingerprint image databases.
5. Methodology of research, results and comparisons
The work carried out in this research is focusing on basic three aspects of retrieval process;
feature extraction, classification and retrieval. The other aspect of the system is the kind of
input images. Standard image database NIST-4 contains the gray scale fingerprint image
pairs taken by a specific acquisition source. FVC2000, FVC2002 and FVC2004 databases
contains eight images of a fingerprint acquired from a specific source. The captured image
using optical sensor with the help of volunteers contains four images for a fingerprint.
5.1 Fingerprint Classification Algorithm - 1
The work proposed for fingerprint classification in this algorithm has major two steps; feature
extraction using LBP and GLCM and classification using Support Vector Machine and k-
Nearest Neighbor classifiers.
5.1.1 Hybrid Feature Extraction using LBP and GLCM
5.1.1.1 Local Binary Pattern (LBP)
Feature extraction is the most important phase for classification and retrieval. The proposed
work use of LBP features due to their strong local feature representation and computational
9 | P a g e
simplicity. LBP features describe changes of ridge patterns of fingerprint images of different
classes and hence, would be efficient and effective for classification. The LBP operator labels
the pixels of an image by thresholding a 3×3 neighborhood of each pixel with the centre
value and considering results as a binary number, called as LBP Codes. After labelling a
image with LBP operator, a histogram of the labelled image is found. This 256-bin histogram
of LBP labels computed over a region contains information about the distribution of the local
micro-patterns, such as edges of the ridges, ridge end-points and ridge bifurcations, so can be
used as a texture descriptor to describe fingerprint image characteristics.
A LBP histogram computed over the whole face image encodes only the occurrences of the
micro-patterns without any indications about their locations. The extracted feature histogram
represents the local texture and global shape of fingerprint images.
5.1.1.2 Gray Level Co-occurrence Matrix (GLCM)
GLCM is a statistical method of examining texture that considers the spatial relationship of
pixels. The GLCM functions create GLC Matrix by characterizing the texture of an image by
calculating how often pairs of pixel with specific values and in a specified spatial relationship
occur in an image and then extract statistical measures from this matrix. Statistics can derived
from this matrix provide information about the texture of an image. Few of them are;
Entropy, Correlation, Energy and Homogeneity.
Entropy is a statistical measure of randomness that can be used to characterize the texture of
the input image. Correlation measures the dependency of grey levels of specified neighboring
pixel pairs. Energy provides the sum of squared elements in the GLCM. It gives high value
when image has very good homogeneity, i.e. when pixels are very similar. Homogeneity
measures the closeness of the distribution of elements in the GLCM to the GLCM diagonal.
The features from LBP and GLCM concatenated to obtain a single vector that is given to the
classifier.
5.1.2 Classification
5.1.2.1 Classification using kNN Classifier
k-Nearest Neighbor classifies data sets based on their similarity with neighbors where „k‟
stands for number of data set items that are considered for the classification. kNN makes
predictions for a new instance(x) by searching through the entire training set for k most
10 | P a g e
similar instances (neighbors). To determine the similarity, a distance measure is used. For
real valued input variables, Euclidian distance is used.
5.1.2.2 Classification using SVM Classifier
Support Vector Machine (SVM) is one of the known algorithms for pattern and image
classification. SVM is designed to separate of the training images of different classes in
feature space. It builds the optimal separating hyper planes based on a kernel function. All
images of which feature vector lies on one side of the hyper plane are belong to one class.
5.2 Fingerprint Classification Algorithm – 2
The work proposed for fingerprint classification in this algorithm has major two steps; feature
extraction using Speed Up Robust Features (SURF) and classification using BoF.
5.2.1 Feature Extraction using SURF
Feature extraction algorithm using SURF selects the interest points from the images at
different scales using Hessian Matrix which can represent the texture of the image uniquely.
In the initial step, the input image is represented as an Integral Image which minimizes the
cost of calculating sum of intensity values of a rectangular region. With the neighborhood of
every interest points, then it builds the feature vector which is distinctive and robust to noise
and geometric deformations. To make features invariant to rotation, Haar wavelet responses
are calculated within the circular neighborhood abound the interest point.
5.2.2 Classification using BoF
A BoF method is one that represents fingerprint images as orderless collections of local
features. In the first step, from the features extracted by SURF technique, it clusters them into
the pair of image and set of features contained in that image, to build a visual vocabulary.
Then it records the count of each clusters appear in the image to create a normalized
histogram representing cluster vector, which is known as BoF. Then it uses Nearest Neighbor
method it assigns the cluster to the fingerprint image to declare the class.
5.3 Fingerprint Retrieval Algorithm
After declaring the class of fingerprint image, the retrieval of similar fingerprint images is
performed within the class declared.
11 | P a g e
5.3.1 Hybrid Feature Extraction using SURF and SIFT
In the first step of SIFT feature extraction, algorithm starts with the scale-space extrema
detection in which it identify the potential interest points by difference of Gaussian function
(DoG). Then it determines the location and scale of interest point by detecting local maxima
and minima of DoG. Interest point orientation can be calculated by orientation of histogram
of local gradients. In the final stage , the local image gradients are measured at the selected
scale in the region around interest point and transformed into scale-invariant coordinates
relative to local features and feature vector is generated.
The features from SIFT and SURF concatenated to obtain a single vector that is given to the
retrieval algorithm.
5.3.2 Retrieval using BoF
As explained earlier, BoF generate the clusters of the hybrid features generated in the
previous step and tries to find the existence of feature cluster in the fingerprint image. Then it
ranks each image based on its distance from the given fingerprint image, calculated based on
Nearest Neighbor strategy. The rank results are ordered and the nearest n images whose
distance shows the nearest images are displayed.
6. Achievements with respect to objectives
The objective of this work was to devise a method for accurate fingerprint image
classification which minimizes the comparisons at retrieval stage. The retrieval method is
expected to be accurate and speedy.
Different methods are proposed to achieve the stated objective. The feature extraction
methods discussed in algorithm 1 are utilizing the advantages of both the methods LBP and
GLCM to improve the classification rate. The SURF method, discussed in algorithm 2,
speeds up the feature extraction process and hence improves the system response time. The
classification is performed on standard databases as well as manually captured images.
The retrieval method proposed for fingerprint retrieval was carried out by utilizing the scale
and rotation invariant features SIFT and SURF. Feature extraction is fast due to the SURF
characteristics. Retrieval results are showing significant improvement using standard as well
as manually captured image databases.
12 | P a g e
7. Conclusion
The feature extraction methods used for classification of fingerprint images are evaluated by
generating confusion matrices by all combinations of four feature extraction methods (LBP,
GLCM, Hybrid and SURF) and three classification methods (SVM, kNN and BoF). The
experiments are performed on NIST-4 fingerprint image database. The experiment results
are cross-validated using k-fold model. The results shown significant improvement when
Hybrid and SURF methods used with BoF classifier.
Further, the fingerprint image retrieval is performed by using Hybrid feature extraction
method and BoF for retrieval. The experiments are carried out on NIST-4 fingerprint image
database and FVC databases by the combinations of three feature extraction methods (SIRF,
SIFT and Hybrid) with BoF. The performance of the system is evaluated based on the
performance measures precision and recall. Due to limited number of images of single
fingerprint images, recall gives better evaluation analysis than precision. The results of
Hybrid feature extraction method with BoF shows improvement in retrieval rate.
13 | P a g e
8. List of publications
[1] Sudhir Vegad, Dr. Tanmay Pawar, "Fingerprint Image Retrieval: A Novel Approach”,
Springer International Conference on Intelligent Systems and Signal Processing, ISSP-
2017, March 2017
[2] Sudhir Vegad, Dr. Tanmay Pawar, “Fingerprint Image Retrieval using Statistical
Methods”, International Journal of Computer Science and Information Security, ISSN –
1947 5500, Vol. 15 (11), pp 246-249, November 2017
14 | P a g e
9. References
[1] Kai Cao, Liaojun Pang, Jimin Liang, Jie Tian, “Fingerprint classification by a
hierarchical classifier”, Pattern Recognition, Elsevier, Vol. 46 (12), pp 3186-3197,
December 2013.
[2] Shing Chyi Chua1, Eng Kiong Wong and Alan Wee Chiat Tan, “A Fuzzy Rule-Based
Fingerprint Image Classification”, International Journal of Applied Engineering
Research ISSN 0973-4562, Vol. 11 (13), pp 7920-7925, 2016.
[3] S. Govindaraju and G. P. Ramesh Kumar, “A Novel Content Based Medical Image
Retrieval using SURF Features”, Indian Journals of Science and Technology, ISSN
0974-6846, Vol 9(20), May 2016.
[4] Nanne van Noord and Eric Postma “Learning Scale-variant and Scale-invariant features
for deep image”, Pattern Recognition, Elsevier, 2017.
[5] Manhua Liu, “Fingerprint classification based on Adaboost learning from singularity
features”, Pattern Recognition, Elsevier, pp 1062-1070, 2010.
[6] Sen Wang, Wei Wei Zhang and Yang Sheng Wang, “Fingerprint Classification by
Directional Fields”, IEEE International Conference on Multimedia Interfaces, 2002.
[7] Wei Zhou, Jiankun Hu, Song Wang and Ian Petersen, “Fingerprint Indexing Based on
Combination of Novel Minutiae Triplet Features”, Network and System Security
Lecture Notes in Computer Science, vol. 8792, pp. 377-388, 2014.
[8] L. H. Thai, T. S. Hai, N. T. Thuy, “Image classification using support vector machine
and artificial neural network”, International Journal of Information Technology and
Computer Science, Vol. 5, pp 32-38, March 2012.
[9] P. Mohanaiah, P. Sathyanarayana and L. GuruKumar, “Image Texture feature
extraction using GLCM approach”, International Journal of Scientific and Research
Publications, ISSN 2250-3153, Volume 3 (5), May 2013.
[10] A Suresh, KL Shunmuganathan, “Image Texture Classification using Gray Level Co-
Occurrence Matrix Based Statistical Features”, European journal of Scientific Research,
Vol.75 (4), pp. 591-597, 2012.
[11] K. C. Leung and C. H. Leung, “Improvement of Fingerprint Retrieval by a Statistical
Classifier”, IEEE Transactions on Information Forensics and Sedurity, Vol. 6 (1),
March 2011.
[12] Matti Pietikäinen, Abdenour Hadid, Guoying Zhao, Timo Ahonen, “Local Binary
Patterns for Still Images”, Computational Imaging and Vision, Springer Book Series,
Vol. 40, pp 13-47, 2011.
15 | P a g e
[13] Fernando Roberti de Siqueira, WilliamRobson Schwartz, HelioPedrini, “Multi-scale
gray level co-occurrence matrices for texture description”, Neurocomputing, Elsevier,
Vol. 120, pp 336-345, November 2013.
[14] Guoying Zhao, Timo Ahonen, Jiří Matas, Matti Pietikainen, “Rotation-Invariant Image
and Video Description With Local Binary Pattern Features”, IEEE Transactions of
Image processing, Vol. 21 (4), pp 1465-1477, November 2011.
[15] Yashika Birdi and Er. Jagroop Kaur, “Wavelet Based Fingerprint Image Retrieval”,
International Journal of Computer Engineering and Applications, ISSN 2321-3469, Vol.
9 (8), September 2015.
[16] Javier A. Montoya Zegarra, Neucimar J. Leite and Ricardo da Silva Torres, “Wavelet-
based fingerprint image retrieval”, Journal of Computational and Applied Mathematics
227, Elsevier, pp 294-307, 2009.
[17] Daniel Peralta, Isaac Triguero, Salvador Garc and Yvan Saeys, “On the use of
convolutional neural networks for robust classification of multiple fingerprint”,
International Journal of Intelligent Systems, November 2017.
[18] Shing Chyi Chua, Eng Kiong Wong and Alan Wee Chiat Tan, “Fingerprint Singular
Point Detection via Quantization and Fingerprint Classification”, World of Computer
Science and Information Technology Journal, ISSN: 2221-0741, Vol. 5 (12), pp 172-
179, 2015.
[19] Anilkumar Muthevi and Dr Ravi Babu Uppu, “Leaf Classification Using Completed
Local Binary Pattern Of Textures”, IEEE 7th International Advance Computing
Conference, 2017.
[20] Fahimeh Alaei, Alireza Alaei, Michael Blumenstein and Umapada Pal, “A Brief
Review of Document Image Retrieval Methods: Recent Advances”, International Joint
Conference on Neural Network, July 2016.
[21] Yani Wang, Zhendong Wu and Jianwu Zhang, “Damaged Fingerprint Classification by
Deep Learning With Fuzzy Feature Points”, International Congress on Image and
Signal Processing, BioMedical Engineering and Informatics, 2016.
[22] Nitesh Chauhan, Mayank Soni, Vijay Anand and Vivek Kanhangad, “Fingerprint
Classification Using Crease Features”, Proceedings of IEEE Students‟ Technology
Symposium, 2016.
[23] Zinelabidine Boulkenafet, Jukka Komulainen and Abdenour Hadid, “Face Antispoofing
Using Speeded-Up Robust Features and Fisher Vector Encoding”, IEEE Signal
Processing Letters, Vol. 24 (2), February 2017.
[24] Nouman Ali, Khalid Bashir Bajwa, Robert Sablatnig, Savvas A. Chatzichristofis,
Zeshan Iqbal, Muhammad Rashid and Hafiz Adnan Habib, “A Novel Image Retrieval
16 | P a g e
Based on Visual Words Integration of SIFT and SURF”, Research Article, PLOS one,
June 2016.
[25] Jinyi Zou, Wei Li, Chen Chen and Qian Du, “Scene classification using local and global
features with collaborative representation fusion”, Information Sciences, Elsevier, pp
209-226, 2016.
[26] Mikel Galar, Joaquín Derrac, Daniel Peralta, Isaac Triguero, Daniel Paternain, Carlos
Lopez-Molina, Salvador García, José M. Benítez, Miguel Pagola, Edurne Barrenechea,
Humberto Bustince and Francisco Herrera, “A survey of fingerprint classification Part
I: Experimental analysis and ensemble proposal”, Knowledge Based Systems, Elsevier,
pp. 76-97, 2015.
[27] Mikel Galar, Joaquín Derrac, Daniel Peralta, Isaac Triguero, Daniel Paternain, Carlos
Lopez-Molina, Salvador García, José M. Benítez, Miguel Pagola, Edurne Barrenechea,
Humberto Bustince and Francisco Herrera, “A survey of fingerprint classification Part
II: Experimental analysis and ensemble proposal”, Knowledge Based Systems, Elsevier,
pp. 98-106, 2015.
[28] Sukhmanjeet Kaur and Mr. Prince Verma, “Content Based Image Retrieval: Integration
of Neural Networks Using Speed-Up Robust Feature and SVM”, International Journal
of Computer Science and Information Technologies, Vol. 6 (1), pp 243-248, 2015.
[29] Kribashnee Dorasamy, Leandra Webb, Jules Tapamo and Nontokozo P. Khanyile,
“Fingerprint classification using a simplified rule-set based on directional patterns and
singularity features”, IEEE International Conference on Biometrics, 2015.
[30] L. H. Thai, T. S. Hai, N. T. Thuy, “Image classification using support vector machine
and artificial neural network”, International Journal of Information Technology and
Computer Science, Vol. 5, pp 32-38, March 2012.
[31] Chang, R. I., Lin, S. Y., Ho, J. M., Fann, C. W., & Wang, Y. C. (2012). “A novel
content based image retrieval system using k-means / knn with feature extraction“.
Computer Science and Information Systems, 9(4), pp.1645-1661. 2012.
[32] Tsai, C. F. “Bag-of-words representation in image annotation: A review”, International
Scholarly Research Network Artificial Intelligence, 2012.
[33] Herbert Bay, Andreas Ess, Tinne Tuytelaars Luc Van Gool, “Speeded-Up Robust
Features (SURF)”, Computer Vision and Image Understanding, Elsevier, Vol. 110 (3),
pp 346-359, 2008.
[34] Hiroharu Kato and Tatsuya Harada, “Image Reconstruction from Bag-of-Visual-
Words”, IEEE Conference on Computer Vision and Pattern Recognition, 2014.
[35] E.N. Mortensen, Hongli Deng and L. Shapiro, “A SIFT descriptor with global context”,
IEEE Computer Society on Computer Vision and Pattern Recognitin, 2005.
17 | P a g e
[36] Ray-I Chang, Shu-Yu Lin, Jan-Ming Ho, Chi-Wen Fann and Yu-Chun Wang, “Facial
Expression Classification Using Artificial Neural Network and K-Nearest Neighbor”,
International Journal of Information Technology and Computer Science, Vol. 3, pp 27-
32, 2015.
[37] Nursabillilah Mohd Ali, Soon Wei Jun, Mohd Safirin Karis, Mariam Md Ghazaly and
Mohd Shahrieel Mohd Aras, “Object classification and recognition using Bag-of-Words
(BoW) model”, IEEE International Colloquium on Signal Processing & Its
Applications, July 2016.