Abstract: In this paper, we present the region based hidden Markov random field model (RBHMRF), which encodes the characteristics of different brain regions into a probabilistic framework for brain MR image segmentation. The recently proposed TV+L1 model is used for region extraction. By utilizing different spatial characteristics in different brain regions, the RMHMRF model performs beyond the current state-of-the-art method, the hidden Markov random field model (HMRF), which uses identical spatial information throughout the whole brain. Experiments on both real and synthetic 3D MR images show that the segmentation result of the proposed method has higher accuracy compared to existing algorithms.
Abstract: Distant-talking voice-based HCI system suffers from
performance degradation due to mismatch between the acoustic
speech (runtime) and the acoustic model (training). Mismatch is
caused by the change in the power of the speech signal as observed at
the microphones. This change is greatly influenced by the change in
distance, affecting speech dynamics inside the room before reaching
the microphones. Moreover, as the speech signal is reflected, its
acoustical characteristic is also altered by the room properties. In
general, power mismatch due to distance is a complex problem. This
paper presents a novel approach in dealing with distance-induced
mismatch by intelligently sensing instantaneous voice power variation
and compensating model parameters. First, the distant-talking speech
signal is processed through microphone array processing, and the
corresponding distance information is extracted. Distance-sensitive
Gaussian Mixture Models (GMMs), pre-trained to capture both
speech power and room property are used to predict the optimal
distance of the speech source. Consequently, pre-computed statistic
priors corresponding to the optimal distance is selected to correct
the statistics of the generic model which was frozen during training.
Thus, model combinatorics are post-conditioned to match the power
of instantaneous speech acoustics at runtime. This results to an
improved likelihood in predicting the correct speech command at
farther distances. We experiment using real data recorded inside two
rooms. Experimental evaluation shows voice recognition performance
using our method is more robust to the change in distance compared
to the conventional approach. In our experiment, under the most
acoustically challenging environment (i.e., Room 2: 2.5 meters), our
method achieved 24.2% improvement in recognition performance
against the best-performing conventional method.
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: Skin color based tracking techniques often assume a
static skin color model obtained either from an offline set of library
images or the first few frames of a video stream. These models
can show a weak performance in presence of changing lighting or
imaging conditions. We propose an adaptive skin color model based
on the Gaussian mixture model to handle the changing conditions.
Initial estimation of the number and weights of skin color clusters
are obtained using a modified form of the general Expectation
maximization algorithm, The model adapts to changes in imaging
conditions and refines the model parameters dynamically using spatial
and temporal constraints. Experimental results show that the method
can be used in effectively tracking of hand and face regions.
Abstract: A state of the art Speaker Identification (SI) system requires a robust feature extraction unit followed by a speaker modeling scheme for generalized representation of these features. Over the years, Mel-Frequency Cepstral Coefficients (MFCC) modeled on the human auditory system has been used as a standard acoustic feature set for SI applications. However, due to the structure of its filter bank, it captures vocal tract characteristics more effectively in the lower frequency regions. This paper proposes a new set of features using a complementary filter bank structure which improves distinguishability of speaker specific cues present in the higher frequency zone. Unlike high level features that are difficult to extract, the proposed feature set involves little computational burden during the extraction process. When combined with MFCC via a parallel implementation of speaker models, the proposed feature set outperforms baseline MFCC significantly. This proposition is validated by experiments conducted on two different kinds of public databases namely YOHO (microphone speech) and POLYCOST (telephone speech) with Gaussian Mixture Models (GMM) as a Classifier for various model orders.
Abstract: One major source of performance decline in speaker
recognition system is channel mismatch between training and testing.
This paper focuses on improving channel robustness of speaker
recognition system in two aspects of channel compensation technique
and channel robust features. The system is text-independent speaker
identification system based on two-stage recognition. In the aspect of
channel compensation technique, this paper applies MAP (Maximum
A Posterior Probability) channel compensation technique, which was
used in speech recognition, to speaker recognition system. In the
aspect of channel robust features, this paper introduces
pitch-dependent features and pitch-dependent speaker model for the
second stage recognition. Based on the first stage recognition to
testing speech using GMM (Gaussian Mixture Model), the system
uses GMM scores to decide if it needs to be recognized again. If it
needs to, the system selects a few speakers from all of the speakers
who participate in the first stage recognition for the second stage
recognition. For each selected speaker, the system obtains 3
pitch-dependent results from his pitch-dependent speaker model, and
then uses ANN (Artificial Neural Network) to unite the 3
pitch-dependent results and 1 GMM score for getting a fused result.
The system makes the second stage recognition based on these fused
results. The experiments show that the correct rate of two-stage
recognition system based on MAP channel compensation technique
and pitch-dependent features is 41.7% better than the baseline system
for closed-set test.
Abstract: Current image-based individual human recognition
methods, such as fingerprints, face, or iris biometric modalities
generally require a cooperative subject, views from certain aspects,
and physical contact or close proximity. These methods cannot
reliably recognize non-cooperating individuals at a distance in the
real world under changing environmental conditions. Gait, which
concerns recognizing individuals by the way they walk, is a relatively
new biometric without these disadvantages. The inherent gait
characteristic of an individual makes it irreplaceable and useful in
visual surveillance.
In this paper, an efficient gait recognition system for human
identification by extracting two features namely width vector of
the binary silhouette and the MPEG-7-based region-based shape
descriptors is proposed. In the proposed method, foreground objects
i.e., human and other moving objects are extracted by estimating
background information by a Gaussian Mixture Model (GMM) and
subsequently, median filtering operation is performed for removing
noises in the background subtracted image. A moving target classification
algorithm is used to separate human being (i.e., pedestrian)
from other foreground objects (viz., vehicles). Shape and boundary
information is used in the moving target classification algorithm.
Subsequently, width vector of the outer contour of binary silhouette
and the MPEG-7 Angular Radial Transform coefficients are taken as
the feature vector. Next, the Principal Component Analysis (PCA)
is applied to the selected feature vector to reduce its dimensionality.
These extracted feature vectors are used to train an Hidden Markov
Model (HMM) for identification of some individuals. The proposed
system is evaluated using some gait sequences and the experimental
results show the efficacy of the proposed algorithm.
Abstract: This paper presents a new approach for image
segmentation by applying Pillar-Kmeans algorithm. This
segmentation process includes a new mechanism for clustering the
elements of high-resolution images in order to improve precision and
reduce computation time. The system applies K-means clustering to
the image segmentation after optimized by Pillar Algorithm. The
Pillar algorithm considers the pillars- placement which should be
located as far as possible from each other to withstand against the
pressure distribution of a roof, as identical to the number of centroids
amongst the data distribution. This algorithm is able to optimize the
K-means clustering for image segmentation in aspects of precision
and computation time. It designates the initial centroids- positions
by calculating the accumulated distance metric between each data
point and all previous centroids, and then selects data points which
have the maximum distance as new initial centroids. This algorithm
distributes all initial centroids according to the maximum
accumulated distance metric. This paper evaluates the proposed
approach for image segmentation by comparing with K-means and
Gaussian Mixture Model algorithm and involving RGB, HSV, HSL
and CIELAB color spaces. The experimental results clarify the
effectiveness of our approach to improve the segmentation quality in
aspects of precision and computational time.