Abstract: The design of a pattern classifier includes an attempt
to select, among a set of possible features, a minimum subset of
weakly correlated features that better discriminate the pattern classes.
This is usually a difficult task in practice, normally requiring the
application of heuristic knowledge about the specific problem
domain. The selection and quality of the features representing each
pattern have a considerable bearing on the success of subsequent
pattern classification. Feature extraction is the process of deriving
new features from the original features in order to reduce the cost of
feature measurement, increase classifier efficiency, and allow higher
classification accuracy. Many current feature extraction techniques
involve linear transformations of the original pattern vectors to new
vectors of lower dimensionality. While this is useful for data
visualization and increasing classification efficiency, it does not
necessarily reduce the number of features that must be measured
since each new feature may be a linear combination of all of the
features in the original pattern vector. In this paper a new approach is
presented to feature extraction in which feature selection, feature
extraction, and classifier training are performed simultaneously using
a genetic algorithm. In this approach each feature value is first
normalized by a linear equation, then scaled by the associated weight
prior to training, testing, and classification. A knn classifier is used to
evaluate each set of feature weights. The genetic algorithm optimizes
a vector of feature weights, which are used to scale the individual
features in the original pattern vectors in either a linear or a nonlinear
fashion. By this approach, the number of features used in classifying
can be finely reduced.
Abstract: An image texture analysis and target recognition approach of using an improved image texture feature coding method (TFCM) and Support Vector Machine (SVM) for target detection is presented. With our proposed target detection framework, targets of interest can be detected accurately. Cascade-Sliding-Window technique was also developed for automated target localization. Application to mammogram showed that over 88% of normal mammograms and 80% of abnormal mammograms can be correctly identified. The approach was also successfully applied to Synthetic Aperture Radar (SAR) and Ground Penetrating Radar (GPR) images for target detection.
Abstract: This paper proposes a novel feature extraction method,
based on Discrete Wavelet Transform (DWT) and K-L Seperability
(KLS), for the classification of Functional Data (FD). This method
combines the decorrelation and reduction property of DWT and the
additive independence property of KLS, which is helpful to extraction
classification features of FD. It is an advanced approach of the
popular wavelet based shrinkage method for functional data reduction
and classification. A theory analysis is given in the paper to prove the
consistent convergence property, and a simulation study is also done
to compare the proposed method with the former shrinkage ones. The
experiment results show that this method has advantages in improving
classification efficiency, precision and robustness.
Abstract: The k-nearest neighbors (knn) is a simple but effective method of classification. In this paper we present an extended version of this technique for chemical compounds used in High Throughput Screening, where the distances of the nearest neighbors can be taken into account. Our algorithm uses kernel weight functions as guidance for the process of defining activity in screening data. Proposed kernel weight function aims to combine properties of graphical structure and molecule descriptors of screening compounds. We apply the modified knn method on several experimental data from biological screens. The experimental results confirm the effectiveness of the proposed method.
Abstract: Support vector machines (SVMs) are considered to be
the best machine learning algorithms for minimizing the predictive
probability of misclassification. However, their drawback is that for
large data sets the computation of the optimal decision boundary is a
time consuming function of the size of the training set. Hence several
methods have been proposed to speed up the SVM algorithm. Here
three methods used to speed up the computation of the SVM
classifiers are compared experimentally using a musical genre
classification problem. The simplest method pre-selects a random
sample of the data before the application of the SVM algorithm. Two
additional methods use proximity graphs to pre-select data that are
near the decision boundary. One uses k-Nearest Neighbor graphs and
the other Relative Neighborhood Graphs to accomplish the task.
Abstract: The tagging data of (users, tags and resources) constitutes a folksonomy that is the user-driven and bottom-up approach to organizing and classifying information on the Web. Tagging data stored in the folksonomy include a lot of very useful information and knowledge. However, appropriate approach for analyzing tagging data and discovering hidden knowledge from them still remains one of the main problems on the folksonomy mining researches. In this paper, we have proposed a folksonomy data mining approach based on FCA for discovering hidden knowledge easily from folksonomy. Also we have demonstrated how our proposed approach can be applied in the collaborative tagging system through our experiment. Our proposed approach can be applied to some interesting areas such as social network analysis, semantic web mining and so on.
Abstract: Shadow detection is still considered as one of the
potential challenges for intelligent automated video surveillance
systems. A pre requisite for reliable and accurate detection and
tracking is the correct shadow detection and classification. In such a
landscape of conditions, privacy issues add more and more
complexity and require reliable shadow detection.
In this work the intertwining between security, accuracy,
reliability and privacy is analyzed and, accordingly, a novel
architecture for Privacy Enhancing Video Surveillance (PEVS) is
introduced. Shadow detection and masking are dealt with through the
combination of two different approaches simultaneously. This results
in a unique privacy enhancement, without affecting security.
Subsequently, the methodology was employed successfully in a
large-scale wireless video surveillance system; privacy relevant
information was stored and encrypted on the unit, without
transferring it over an un-trusted network.
Abstract: An improved processing description to be employed in biosonar signal processing in a cochlea model is proposed and examined. It is compared to conventional models using a modified discrimination analysis and both are tested. Their performances are evaluated with echo data captured from natural targets (trees).Results indicate that the phase characteristics of low-pass filters employed in the echo processing have a significant effect on class separability for this data.
Abstract: Mammography is the most effective procedure for an
early diagnosis of the breast cancer. Nowadays, people are trying to
find a way or method to support as much as possible to the
radiologists in diagnosis process. The most popular way is now being
developed is using Computer-Aided Detection (CAD) system to
process the digital mammograms and prompt the suspicious region to
radiologist. In this paper, an automated CAD system for detection
and classification of massive lesions in mammographic images is
presented. The system consists of three processing steps: Regions-Of-
Interest detection, feature extraction and classification. Our CAD
system was evaluated on Mini-MIAS database consisting 322
digitalized mammograms. The CAD system-s performance is
evaluated using Receiver Operating Characteristics (ROC) and Freeresponse
ROC (FROC) curves. The archived results are 3.47 false
positives per image (FPpI) and sensitivity of 85%.
Abstract: Text categorization is the problem of classifying text
documents into a set of predefined classes. After a preprocessing
step, the documents are typically represented as large sparse vectors.
When training classifiers on large collections of documents, both the
time and memory restrictions can be quite prohibitive. This justifies
the application of feature selection methods to reduce the
dimensionality of the document-representation vector. In this paper,
three feature selection methods are evaluated: Random Selection,
Information Gain (IG) and Support Vector Machine feature selection
(called SVM_FS). We show that the best results were obtained with
SVM_FS method for a relatively small dimension of the feature
vector. Also we present a novel method to better correlate SVM
kernel-s parameters (Polynomial or Gaussian kernel).
Abstract: This paper proposes a new approach to perform the
problem of real-time face detection. The proposed method combines
primitive Haar-Like feature and variance value to construct a new
feature, so-called Variance based Haar-Like feature. Face in image
can be represented with a small quantity of features using this
new feature. We used SVM instead of AdaBoost for training and
classification. We made a database containing 5,000 face samples
and 10,000 non-face samples extracted from real images for learning
purposed. The 5,000 face samples contain many images which have
many differences of light conditions. And experiments showed that
face detection system using Variance based Haar-Like feature and
SVM can be much more efficient than face detection system using
primitive Haar-Like feature and AdaBoost. We tested our method on
two Face databases and one Non-Face database. We have obtained
96.17% of correct detection rate on YaleB face database, which is
higher 4.21% than that of using primitive Haar-Like feature and
AdaBoost.
Abstract: One of the main environmental problems which affect extensive areas in the world is soil salinity. Traditional data collection methods are neither enough for considering this important environmental problem nor accurate for soil studies. Remote sensing data could overcome most of these problems. Although satellite images are commonly used for these studies, however there are still needs to find the best calibration between the data and real situations in each specified area. Neyshaboor area, North East of Iran was selected as a field study of this research. Landsat satellite images for this area were used in order to prepare suitable learning samples for processing and classifying the images. 300 locations were selected randomly in the area to collect soil samples and finally 273 locations were reselected for further laboratory works and image processing analysis. Electrical conductivity of all samples was measured. Six reflective bands of ETM+ satellite images taken from the study area in 2002 were used for soil salinity classification. The classification was carried out using common algorithms based on the best composition bands. The results showed that the reflective bands 7, 3, 4 and 1 are the best band composition for preparing the color composite images. We also found out, that hybrid classification is a suitable method for identifying and delineation of different salinity classes in the area.
Abstract: A new conceptual architecture for low-level neural
pattern recognition is presented. The key ideas are that the brain
implements support vector machines and that support vectors are
represented as memory patterns in competitive queuing memories. A
binary classifier is built from two competitive queuing memories
holding positive and negative valence training examples respectively.
The support vector machine classification function is calculated in
synchronized evaluation cycles. The kernel is computed by bisymmetric
feed-forward networks feed by sensory input and by
competitive queuing memories traversing the complete sequence of
support vectors. Temporary summation generates the output
classification. It is speculated that perception apparatus in the brain
reuses structures that have evolved for enabling fluent execution of
prepared action sequences so that pattern recognition is built on
internalized motor programmes.
Abstract: In this paper, we present a new and effective image indexing technique that extracts features directly from DCT domain. Our proposed approach is an object-based image indexing. For each block of size 8*8 in DCT domain a feature vector is extracted. Then, feature vectors of all blocks of image using a k-means algorithm is clustered into groups. Each cluster represents a special object of the image. Then we select some clusters that have largest members after clustering. The centroids of the selected clusters are taken as image feature vectors and indexed into the database. Also, we propose an approach for using of proposed image indexing method in automatic image classification. Experimental results on a database of 800 images from 8 semantic groups in automatic image classification are reported.
Abstract: Wavelet transform provides several important
characteristics which can be used in a texture analysis and
classification. In this work, an efficient texture classification method,
which combines concepts from wavelet and co-occurrence matrices,
is presented. An Euclidian distance classifier is used to evaluate the
various methods of classification. A comparative study is essential to
determine the ideal method. Using this conjecture, we developed a
novel feature set for texture classification and demonstrate its
effectiveness
Abstract: Classification is one of the primary themes in
computational biology. The accuracy of classification strongly
depends on quality of a dataset, and we need some method to
evaluate this quality. In this paper, we propose a new graphical
analysis method using 'Membership-Deviation Graph (MDG)' for
analyzing quality of a dataset. MDG represents degree of
membership and deviations for instances of a class in the dataset. The
result of MDG analysis is used for understanding specific feature and
for selecting best feature for classification.
Abstract: In this paper in consideration of each available
techniques deficiencies for speech recognition, an advanced method
is presented that-s able to classify speech signals with the high
accuracy (98%) at the minimum time. In the presented method, first,
the recorded signal is preprocessed that this section includes
denoising with Mels Frequency Cepstral Analysis and feature
extraction using discrete wavelet transform (DWT) coefficients; Then
these features are fed to Multilayer Perceptron (MLP) network for
classification. Finally, after training of neural network effective
features are selected with UTA algorithm.
Abstract: An unsupervised classification algorithm is derived
by modeling observed data as a mixture of several mutually
exclusive classes that are each described by linear combinations of
independent non-Gaussian densities. The algorithm estimates the
data density in each class by using parametric nonlinear functions
that fit to the non-Gaussian structure of the data. This improves
classification accuracy compared with standard Gaussian mixture
models. When applied to textures, the algorithm can learn basis
functions for images that capture the statistically significant structure
intrinsic in the images. We apply this technique to the problem of
unsupervised texture classification and segmentation.
Abstract: We provide a supervised speech-independent voice recognition technique in this paper. In the feature extraction stage we propose a mel-cepstral based approach. Our feature vector classification method uses a special nonlinear metric, derived from the Hausdorff distance for sets, and a minimum mean distance classifier.
Abstract: In modern human computer interaction systems
(HCI), emotion recognition is becoming an imperative characteristic.
The quest for effective and reliable emotion recognition in HCI has
resulted in a need for better face detection, feature extraction and
classification. In this paper we present results of feature space analysis
after briefly explaining our fully automatic vision based emotion
recognition method. We demonstrate the compactness of the feature
space and show how the 2d/3d based method achieves superior features
for the purpose of emotion classification. Also it is exposed that
through feature normalization a widely person independent feature
space is created. As a consequence, the classifier architecture has
only a minor influence on the classification result. This is particularly
elucidated with the help of confusion matrices. For this purpose
advanced classification algorithms, such as Support Vector Machines
and Artificial Neural Networks are employed, as well as the simple k-
Nearest Neighbor classifier.