Face recognition in constrained environment using hidden Markov model approach

Hidden Markov Model(HMM) has a doubly embedded stochastic process with an underlying stochastic process that can be observed through another set of stochastic processes. This structure of HMM is useful for modeling vector sequence that doesn't look like a stochastic process but has a hidden stochastic process. So, HMM approach has become popular in various areas in last decade. The increasing popularity of HMM is based on two facts : rich mathematical structure and proven accuracy on critical application. In this paper, we applied continuous HMM (CHMM) approach with AR coefficient to detect and predict the chatter of lathe bite and to diagnose the wear of oil Journal bearing using rotor shaft displacement. Our examples show that CHMM approach is very efficient method for machine health monitoring and prediction.

In the past years, Hidden Markov Models have been used in several fields and applications with success. More recently, these models have been applied to improve the reliability of a machinery system. In many cases, failure is preceded by specific sequences of events (signature), which can be detected by an adequate Hidden Markov Model (HMM). Classical laws like lifetime models or survival functions are used to estimate the lifetime of a system. The default of these laws is that only the elapsed time is used to estimate the end of life of a system. The aim of this paper is to validate an HMM approach. We first use a synthetic HMM model of degradation to produce event sequences. This synthetic model has been inspired by a real process. In this case, we can adjust the failure rate by changing model parameters. All the parameters of this synthetic model are known and provide references which can be evaluated by different indicators. Classical survival functions used in reliability are computed on synthetic sequences. These laws validate the behavior of the synthetic model. The higher the failure rate, the shorter the lifetime duration. These results confirm that a four-state, left to right, HMM topology can represent the degradation level of a system. In a second time, this HMM approach is used in a real case, where degradation levels are unknown. Degradation estimates are compared with the results from classical survival functions used in the first case. Then we show that the degradation level provided by the HMM approach is more efficient than the survival functions approach. The HMM approach takes into account the events collected about a system, not only the elapsed time as is the case with survival functions.

This paper introduces two unsupervised learning methods for analyzing functional magnetic resonance imaging (fMRI) data based on hidden Markov model (HMM). HMM approach is focused on capturing the first-order statistical evolution among the samples of a voxel time series, and it can provide a complimentary perspective of the BOLD signals. Two-state HMM is created for each voxel, and the model parameters are estimated from the voxel time series and the stimulus paradigm. Two different activation detection methods are presented in this paper. The first method is based on the likelihood and likelihood-ratio test, in which an additional Gaussian model is used to enhance the contrast of the HMM likelihood map. The second method is based on certain distance measures between the two state distributions, in which the most likely HMM state sequence is estimated through the Viterbi algorithm. The distance between the on-state and off-state distributions is measured either through a t-test, or using the Kullback-Leibler distance (KLD). Experimental results on both normal subject and brain tumor subject are presented. HMM approach appears to be more robust in detecting the supplemental active voxels comparing with SPM, especially for brain tumor subject.

We present a path planning framework that takes into account the human's safety perception in the presence of a flying robot. The framework addresses two objectives: (i) estimation of the uncertain parameters of the proposed safety perception model based on test data collected using Virtual Reality (VR) testbed, and (ii) offline optimal control computation using the estimated safety perception model. Due to the unknown factors in the human tests data, it is not suitable to use standard regression techniques that minimize the mean squared error (MSE). We propose to use a Hidden Markov model (HMM) approach where human's attention is considered as a hidden state to infer whether the data samples are relevant to learn the safety perception model. The HMM approach improved log-likelihood over the standard least squares solution. For path planning, we use Bernstein polynomials for discretization, as the resulting path remains within the convex hull of the control points, providing guarantees for deconfliction with obstacles at low computational cost. An example of optimal trajectory generation using the learned human model is presented. The optimal trajectory generated using the proposed model results in reasonable safety distance from the human. In contrast, the paths generated using the standard regression model have undesirable shapes due to overfitting. The example demonstrates that the HMM approach has robustness to the unknown factors compared to the standard MSE model.

Sleep quantity affects an individual's personal health. The gold standard of measuring sleep and diagnosing sleep disorders is Polysomnography (PSG). Although PSG is accurate, it is expensive and it lacks portability. A number of wearable devices with embedded sensors have emerged in the recent past as an alternative to PSG for regular sleep monitoring directly by the user. These devices are intrusive and cause discomfort besides being expensive. In this work, we present an algorithm to detect sleep using a smartphone with the help of its inbuilt accelerometer sensor. We present three different approaches to classify raw acceleration data into two states - Sleep and Wake. In the first approach, we take an equation from Kushida's algorithm to process accelerometer data. Henceforth, we call it Kushida's equation. While the second is based on statistical functions, the third is based on Hidden Markov Model (HMM) training. Although all the three approaches are suitable for a phone's resources, each approach demands different amount of resources. While Kushida's equation-based approach demands the least, the HMM training-based approach demands the maximum. We collected data from mobile phone's accelerometer for four subjects for twelve days each. We compare accuracy of sleep detection using each of the three approaches with that of Zeo sensor, which is based on Electroencephalogram (EEG) sensor to detect sleep. EEG is an important modality in PSG. We find that HMM training-based approach is as much as 84% accurate. It is 15% more accurate as compared to Kushida's equation-based approach and 10% more accurate as compared to statistical method-based approach. In order to concisely represent the sleep quality of people, we model their sleep data using HMM. We present an analysis to find out a tradeoff between the amount of training data and the accuracy provided in the modeling of sleep. We find that six days of sleep data is sufficient for accurate modeling. We compare accuracy of our HMM training-based algorithm with a representative third party app SleepTime available from Google Play Store for Android. We find that the detection done using HMM approach is closer to that done by Zeo by 13% as compared to the third party Android application SleepTime. We show that our HMM training-based approach is efficient as it takes less than ten seconds to get executed on Moto G Android phone.

Masquerade detection is the ability to detect attackers known as masqueraders that intrude on another user’s system and pose as legitimate users. Once a masquerader obtains access to a user’s system, the masquerader has free reign over whatever data is on that system. In this research, we focus on masquerade detection and user classi cation using the following two di erent approaches: the heavy hitter approach and 2 di erent approaches based on hidden Markov models (HMMs), the dueling-HMM and threshold-HMM strategies. The heavy hitter approach computes the frequent elements seen in the training data sequence and test data sequence and computes the distance to see whether the test data sequence is masqueraded or not. The results show very misleading classi cations, suggesting that the approach is not viable for masquerade detection. A hidden Markov model is a tool for representing probability distributions over sequences of observations [9]. Previous research has shown that using a threshold-based hidden Markov model (HMM) approach is successful in a variety of categories: malware detection, intrusion detection, pattern recognition, etc. We have veri ed that using a threshold-based HMM approach produces high accuracy with low amounts of a false positives. Using the dueling- HMM approach, which utilizes multiple training HMMs, we obtain an overall accuracy of 81.96%. With the introduction of the bias in the dueling-HMM approach, we produce similar results to the results obtained in the threshold-based HMM approach, where we see many non-masqueraded data detected, while many masqueraded data avoid detection, yet still result in an high overall accuracy.

<p>This Paper proposes a new facial feature extraction approach, Wash-Hadamard Transform (WHT). This approach is based on correlation between local pixels of the face image. Its primary advantage is the simplicity of its computation. The paper compares the proposed approach, WHT, which was traditionally used in data compression with two other known approaches: the Principal Component Analysis (PCA) and the Discrete Cosine Transform (DCT) using the face database of Olivetti Research Laboratory (ORL). In spite of its simple computation, the proposed algorithm (WHT) gave very close results to those obtained by the PCA and DCT. This paper initiates the research into WHT and the family of frequency transforms and examines their suitability for feature extraction in face recognition applications.</p>

This paper presents a comparison with the conventional watermarking technique and the novel 5-stage pipelined implementation of DCT/IDCT which is used in digital image watermarking. The most common method of Discrete Cosine Transform (DCT)-based digital image watermarking which is used for image authentication and copyright protection is the transpose method. In this method the 2-Dimensional DCT is obtained by taking two 1-dimensional DCTs in series. The image pixel value is first divided into 8x8 blocks and the row-wise 1D DCT of each block is taken. The transpose of the blocks is then determined and a column-wise 1D DCT is ascertained which gives the 2D DCT of the data. The major advantage of this design is that, unlike the conventional DCT-based watermarking technique, this method uses a 5-stage pipeline which can bring about a speed increase of close to 500% over the conventional method which is naturally a great advantage. This technique has been tested on the standard ‘Lena’ image. Both visible and invisible watermarking is implemented in hardware. The design is done in Verilog HDL and the simulation is done in Modelsim 6.3b. Matlab is used to produce the binary data file which is the input to the 1D DCT module. The hardware implementation is done in Xilinx XC3S4000 FPGA. The results of the comparison are discussed in the concluding sections.

Face recognition system usually consists of components of feature extraction and pattern classification. However, not all of extracted facial features contribute to the classification phase positively because of the variations of illumination and poses in face images. In this paper, a three-step feature selection algorithm is proposed in which discrete cosine transform (DCT) and genetic algorithms (GAs) as well as dimensionality reduction methods are utilized to create a combined framework of feature acquisition. In details, the face images are first transformed to frequency domain through DCT. Then GAs are used to seek for optimal features in the redundant DCT coefficients where the generalization performance guides the searching process. The last step is to reduce the dimension of selected features. In experiments, two face databases are used to evaluate the effectiveness of the proposed method. In addition, an entropy-based improvement is also proposed. The experimental results present the superiority of selected frequency features.

In spite of the differences in visual stimulus of human beings such as ageing, changing conditions of a person, and occlusion, recognition can even be done at a glance by the human eye many years after the previous encounter. It has been established that facial differences like the hairstyle changes, growing of one’s beard, wearing of glasses, and other forms of occlusions can hardly hinder the power of the human brain from making a face recognition. However, the same cannot easily be said about automated intelligent systems which have been developed to mimic the skill of the human brain to aid in recognition. There have been growing interests in developing a resilient and efficient recognition system mainly because of its numerous application areas (access control, entertainment/leisure, security system based on biometric data, and user-friendly human-machine interfaces). Although there have been numerous researches on face recognition under varying pose, illumination, expression, and image degradations, problems caused by occlusions are mostly ignored. This study thus focuses on facial occlusions and proposes an enhancement mechanism through face image augmentation to improve the recognition of occluded face images. This study assessed the performance of Principal Component Analysis with Singular Value Decomposition using Fast Fourier Transform (FFT-PCA/SVD) for preprocessing face recognition algorithm on face images with missingness and augmented face image database. It was found that the average recognition rates for the FFT-PCA/SVD algorithm were the same ( 90 % ) when face images with missingness and augmented face images were used as test images, respectively. The statistical evaluation revealed that there exists a significant difference in the average recognition distances for the face images with missingness and augmented face images when FFT-PCA/SVD is used for recognition. Augmented face images tend to have a relatively lower average recognition distance when used as test images. This finding is contrary to the equal performance assessment by the adopted numerical technique. The MICE algorithm is therefore recommended as a suitable imputation mechanism for enhancing/improving the performance of the face recognition system.

In this paper, we report a comprehensive study of two well evolved and developed learning algorithms for effective Face Recognition (FR); viz. the Hidden Markov Model (HMM) and Support Vector Machines (SVM). It is evident that, the accuracy of recognition and efficiency in terms of time and speed of a FR system are directly proportional to the competency of the underlying learning algorithms. Here, we propose to compare the two acclaimed method stated above. A component-based approach is adapted to train face images for recognition. Each face image is divided into sub-states for both HMM and SVM algorithm. In attempt to achieve better rates of recognition, all face images are pre-processed and resizing which helps in reducing the overall complexity of the FR system. We run this proposed system against benchmarks accredited by previous researches.

The statistical appearance of the face can vary due to various factors such as pose, occlusion, expression and background which makes it a challenging task to have an efficient Face Recognition (FR) system. This paper proposes 4 novel techniques viz., Entropy based Face Segregation (EFS) as pre-processing technique, Double Wavelet Noise Removal (DWNR) as pre-processing technique, 1D Stationary Wavelet Transform (SWT) as Feature Extractor and Conservative Binary Particle Optimization (CBPSO) as Feature Selector to enhance the performance of the system. EFS is used to segregate the facial region, thus removing the cluttered background. DWNR has unique combination of 2D Discrete Wavelet Transform (DWT), Wiener Filter and 2D SWT for image denoising and contrast enhancement. The pre-processed image is then fed to unique combination of 1D DWT, 1D SWT and 1D Discrete Cosine Transform (DCT) to extract essential features. CBPSO is used to select very optimum feature subset and significantly reduce the computation time. The proposed algorithm is experimented on four benchmark databases viz., Color FERET, CMU PIE, Pointing Head Pose and Georgia Tech.

This paper proposes an integrated system for face recognition in cluttered images. An extension of the system is able to recognize faces by directly working with JPEG compressed image data without decompressing. The system consists of face detection using a polynomial neural network (PNN) and face recognition using pseudo-2D hidden Markov models (P2D HMM or pseudo-2D HMM). Firstly, images from a database or a Website are rescaled and preprocessed to form a standard size and file format. Secondly, face regions in image are located and extracted by PNN. Finally through face recognition the detected faces are labeled as one person of the database or rejected as new subject. Our experiments show that the face detector has a higher detection rate and low false positive rate. At the same time, the face recognition achieves higher performance with lower complexity and less computation time compared with other systems.

A comprehensive account of almost all the feature extraction methods used in current face recognition systems is presented. Specifically we have made distinction in the holistic and local feature extraction and differentiate them qualitatively as opposed to quantitatively. It is argued that a global feature representation should be preferred over a bag-of-feature approach. The problems in current feature extraction techniques and their reliance on a strict alignment is discussed. Finally we have introduced to use face-GLOH signatures that are invariant with respect to scale, translation and rotation and therefore do not require properly aligned images. The resulting dimensionality of the vector is also low as compared to other commonly used local features such as Gabor, LBP etc. and therefore learning based methods can also benefit from it. In a typical multi-view face recognition task, where it is assumed to have several examples of a subject available for training, we have shown in an extensive experimental setting the advantages and weaknesses of commonly used feature descriptions. Our results show that under more realistic assumptions, most of the classifiers failed on conventional features. While using the introduced face-GLOH-signature representation is relatively less affected by large in-class variations. This has been demonstrated by providing a fair performance comparison of several classifiers under more practical conditions such as misalignments, large number of subjects and large pose variations. An important conclusion is to be drawn from the results on FERET is that conventional multi-view face recognition cannot cope well with regards to large pose variations. Even using a large number of training examples in different poses for a subject do not suffice for a satisfactory recognition. In order to solve the problem where only one training example per subject is available, many recent methods propose to use image synthesis to generate a given subject at all other views and then perform a conventional multi-view recognition (Beymer and Poggio, 1995; Gross et al, 2004). Besides the fact that such synthesis techniques cause severe artefacts and thus cannot preserve the identity of an individual, a conventional classification cannot yield good recognition results, as has been shown in an extensive experimental setting. More sophisticated methods are therefore needed in order to address pose invariant face recognition. Large pose differences cause significant appearance variations that in general are larger than the appearance variation due to identity. One possible way of addressing this is to learn these variations across each pose, more specifically by fixing the pose and establishing a correspondence on how a person’s appearance changes under this pose one could reduce the in-class appearance variation significantly. In our very recent work (Sarfraz and Hellwich, 2009), we demonstrate the usefulness of face-GLOH signature in this direction.

This paper presents a novel illumination normalization approach for face recognition under varying lighting conditions. In the proposed approach, a discrete cosine transform (DCT) is employed to compensate for illumination variations in the logarithm domain. Since illumination variations mainly lie in the low-frequency band, an appropriate number of DCT coefficients are truncated to minimize variations under different lighting conditions. Experimental results on the Yale B database and CMU PIE database show that the proposed approach improves the performance significantly for the face images with large illumination variations. Moreover, the advantage of our approach is that it does not require any modeling steps and can be easily implemented in a real-time face recognition system.