Exploring Transfer Learning Approaches for Head Pose Classification from Multi-view Surveillance Images

Head pose classification from surveillance images acquired with distant, large field-of-view cameras is difficult as faces are captured at low-resolution and have a blurred appearance. Domain adaptation approaches are useful for transferring knowledge from the training (source) to the test (target) data when they have different attributes, minimizing target data labeling efforts in the process. This paper examines the use of transfer learning for efficient multi-view head pose classification with minimal target training data under three challenging situations: (i) where the range of head poses in the source and target images is different, (ii) where source images capture a stationary person while target images capture a moving person whose facial appearance varies under motion due to changing perspective, scale and (iii) a combination of (i) and (ii). On the whole, the presented methods represent novel transfer learning solutions employed in the context of multi-view head pose classification. We demonstrate that the proposed solutions considerably outperform the state-of-the-art through extensive experimental validation. Finally, the DPOSE dataset compiled for benchmarking head pose classification performance with moving persons, and to aid behavioral understanding applications is presented in this work.

Monte Carlo simulation of light transport in turbid medium with embedded object-spherical, cylindrical, ellipsoidal, or cuboidal objects embedded within multilayered tissues

Monte Carlo modeling of light transport in multilayered tissue (MCML) is modified to incorporate objects of various shapes (sphere, ellipsoid, cylinder, or cuboid) with a refractive-index mismatched boundary. These geometries would be useful for modeling lymph nodes, tumors, blood vessels, capillaries, bones, the head, and other body parts. Mesh-based Monte Carlo (MMC) has also been used to compare the results from the MCML with embedded objects (MCML-EO). Our simulation assumes a realistic tissue model and can also handle the transmission/reflection at the object-tissue boundary due to the mismatch of the refractive index. Simulation of MCML-EO takes a few seconds, whereas MMC takes nearly an hour for the same geometry and optical properties. Contour plots of fluence distribution from MCML-EO and MMC correlate well. This study assists one to decide on the tool to use for modeling light propagation in biological tissue with objects of regular shapes embedded in it. For irregular inhomogeneity in the model (tissue), MMC has to be used. If the embedded objects (inhomogeneity) are of regular geometry (shapes), then MCML-EO is a better option, as simulations like Raman scattering, fluorescent imaging, and optical coherence tomography are currently possible only with MCML. (C) 2014 Society of Photo-Optical Instrumentation Engineers (SPIE)

A radial differential mobility analyzer for the size-classification of gas-phase synthesized nanoparticles at low pressures

Differential mobility analyzers (DMAs) are commonly used to generate monodisperse nanoparticle aerosols. Commercial DMAs operate at quasi-atmospheric pressures and are therefore not designed to be vacuum-tight. In certain particle synthesis methods, the use of a vacuum-compatible DMA is a requirement as a process step for producing high-purity metallic particles. A vacuum-tight radial DMA (RDMA) has been developed and tested at low pressures. Its performance has been evaluated by using a commercial NANO-DMA as the reference. The performance of this low-pressure RDMA (LP-RDMA) in terms of the width of its transfer function is found to be comparable with that of other NANO-DMAs at atmospheric pressure and is almost independent of the pressure down to 30 mbar. It is shown that LP-RDMA can be used for the classification of nanometer-sized particles (5-20 nm) under low pressure condition (30 mbar) and has been successfully applied to nanoparticles produced by ablating FeNi at low pressures.

Phylogenetic relationships and classification of thiolases and thiolase-like proteins of Mycobacterium tuberculosis and Mycobacterium smegmatis

Thiolases are enzymes involved in lipid metabolism. Thiolases remove the acetyl-CoA moiety from 3-ketoacyl-CoAs in the degradative reaction. They can also catalyze the reverse Claisen condensation reaction, which is the first step of biosynthetic processes such as the biosynthesis of sterols and ketone bodies. In human, six distinct thiolases have been identified. Each of these thiolases is different from the other with respect to sequence, oligomeric state, substrate specificity and subcellular localization. Four sequence fingerprints, identifying catalytic loops of thiolases, have been described. In this study genome searches of two mycobacterial species (Mycobacterium tuberculosis and Mycobacterium smegmatis), were carried out, using the six human thiolase sequences as queries. Eight and thirteen different thiolase sequences were identified in M. tuberculosis and M. smegmatis, respectively. In addition, thiolase-like proteins (one encoded in the Mtb and two in the Msm genome) were found. The purpose of this study is to classify these mostly uncharacterized thiolases and thiolase-like proteins. Several other sequences obtained by searches of genome databases of bacteria, mammals and the parasitic protist family of the Trypanosomatidae were included in the analysis. Thiolase-like proteins were also found in the trypanosomatid genomes, but not in those of mammals. In order to study the phylogenetic relationships at a high confidence level, additional thiolase sequences were included such that a total of 130 thiolases and thiolase-like protein sequences were used for the multiple sequence alignment. The resulting phylogenetic tree identifies 12 classes of sequences, each possessing a characteristic set of sequence fingerprints for the catalytic loops. From this analysis it is now possible to assign the mycobacterial thiolases to corresponding homologues in other kingdoms of life. The results of this bioinformatics analysis also show interesting differences between the distributions of M. tuberculosis and M. smegmatis thiolases over the 12 different classes. (C) 2014 Elsevier Ltd. All rights reserved.

Diffusing-wave spectroscopy in an inhomogeneous object: Local viscoelastic spectra from ultrasound-assisted measurement of correlation decay arising from the ultrasound focal volume

We demonstrate diffusing-wave spectroscopy (DWS) in a localized region of a viscoelastically inhomogeneous object by measurement of the intensity autocorrelation g(2)(tau)] that captures only the decay introduced by the temperature-induced Brownian motion in the region. The region is roughly specified by the focal volume of an ultrasound transducer which introduces region specific mechanical vibration owing to insonification. Essential characteristics of the localized non-Markovian dynamics are contained in the decay of the modulation depth M(tau)], introduced by the ultrasound forcing in the focal volume selected, on g(2)(tau). The modulation depth M(tau(i)) at any delay time tau(i) can be measured by short-time Fourier transform of g(2)(tau) and measurement of the magnitude of the spectrum at the ultrasound drive frequency. By following the established theoretical framework of DWS, we are able to connect the decay in M(tau) to the mean-squared displacement (MSD) of scattering centers and the MSD to G*(omega), the complex viscoelastic spectrum. A two-region composite polyvinyl alcohol phantom with different viscoelastic properties is selected for demonstrating local DWS-based recovery of G*(omega) corresponding to these regions from the measured region specific M(tau(i))vs tau(i). The ultrasound-assisted measurement of MSD is verified by simulating, using a generalized Langevin equation (GLE), the dynamics of the particles in the region selected as well as by the usual DWS experiment without the ultrasound. It is shown that whereas the MSD obtained by solving the GLE without the ultrasound forcing agreed with its experimental counterpart covering small and large values of tau, the match was good only in the initial transients in regard to experimental measurements with ultrasound.

CLAP: A web-server for automatic classification of proteins with special reference to multi-domain proteins

Background: The function of a protein can be deciphered with higher accuracy from its structure than from its amino acid sequence. Due to the huge gap in the available protein sequence and structural space, tools that can generate functionally homogeneous clusters using only the sequence information, hold great importance. For this, traditional alignment-based tools work well in most cases and clustering is performed on the basis of sequence similarity. But, in the case of multi-domain proteins, the alignment quality might be poor due to varied lengths of the proteins, domain shuffling or circular permutations. Multi-domain proteins are ubiquitous in nature, hence alignment-free tools, which overcome the shortcomings of alignment-based protein comparison methods, are required. Further, existing tools classify proteins using only domain-level information and hence miss out on the information encoded in the tethered regions or accessory domains. Our method, on the other hand, takes into account the full-length sequence of a protein, consolidating the complete sequence information to understand a given protein better. Results: Our web-server, CLAP (Classification of Proteins), is one such alignment-free software for automatic classification of protein sequences. It utilizes a pattern-matching algorithm that assigns local matching scores (LMS) to residues that are a part of the matched patterns between two sequences being compared. CLAP works on full-length sequences and does not require prior domain definitions. Pilot studies undertaken previously on protein kinases and immunoglobulins have shown that CLAP yields clusters, which have high functional and domain architectural similarity. Moreover, parsing at a statistically determined cut-off resulted in clusters that corroborated with the sub-family level classification of that particular domain family. Conclusions: CLAP is a useful protein-clustering tool, independent of domain assignment, domain order, sequence length and domain diversity. Our method can be used for any set of protein sequences, yielding functionally relevant clusters with high domain architectural homogeneity. The CLAP web server is freely available for academic use at http://nslab.mbu.iisc.ernet.in/clap/.

A model for the classification and survey of clock synchronization protocols in WSNs

Clock synchronization in wireless sensor networks (WSNs) assures that sensor nodes have the same reference clock time. This is necessary not only for various WSN applications but also for many system level protocols for WSNs such as MAC protocols, and protocols for sleep scheduling of sensor nodes. Clock value of a node at a particular instant of time depends on its initial value and the frequency of the crystal oscillator used in the sensor node. The frequency of the crystal oscillator varies from node to node, and may also change over time depending upon many factors like temperature, humidity, etc. As a result, clock values of different sensor nodes diverge from each other and also from the real time clock, and hence, there is a requirement for clock synchronization in WSNs. Consequently, many clock synchronization protocols for WSNs have been proposed in the recent past. These protocols differ from each other considerably, and so, there is a need to understand them using a common platform. Towards this goal, this survey paper categorizes the features of clock synchronization protocols for WSNs into three types, viz, structural features, technical features, and global objective features. Each of these categories has different options to further segregate the features for better understanding. The features of clock synchronization protocols that have been used in this survey include all the features which have been used in existing surveys as well as new features such as how the clock value is propagated, when the clock value is propagated, and when the physical clock is updated, which are required for better understanding of the clock synchronization protocols in WSNs in a systematic way. This paper also gives a brief description of a few basic clock synchronization protocols for WSNs, and shows how these protocols fit into the above classification criteria. In addition, the recent clock synchronization protocols for WSNs, which are based on the above basic clock synchronization protocols, are also given alongside the corresponding basic clock synchronization protocols. Indeed, the proposed model for characterizing the clock synchronization protocols in WSNs can be used not only for analyzing the existing protocols but also for designing new clock synchronization protocols. (C) 2014 Elsevier B.V. All rights reserved.

Exploring Technology Evolution Using Patent Classification: A Case of Cochlear Implant Technology Patents

Understanding technology evolution through periodic landscaping is an important stage of strategic planning in R&D Management. In fields like that of healthcare, where the initial R&D investment is huge and good medical product serve patients better, these activities become crucial. Approximately five percentage of the world population has hearing disabilities. Current hearing aid products meet less than ten percent of the global needs. Patent data and classifications on cochlear implants from 1977-2010, show the landscapes and evolution in the area of such implant. We attempt to highlight emergence and disappearance of patent classes over period of time showing variations in cochlear implant technologies. A network analysis technique is used to explore and capture technology evolution in patent classes showing what emerged or disappeared over time. Dominant classes are identified. The sporadic influence of university research in cochlear implants is also discussed.

Experimentally validated Raman Monte Carlo simulation for a cuboid object to obtain Raman spectroscopic signatures for hidden material

In conventional Raman spectroscopic measurements of liquids or surfaces the preferred geometry for detection of the Raman signal is the backscattering (or reflection) mode. For non-transparent layered materials, sub-surface Raman signals have been retrieved using spatially offset Raman spectroscopy (SORS), usually with light collection in the same plane as the point of excitation. However, as a result of multiple scattering in a turbid medium, Raman photons will be emitted in all directions. In this study, Monte Carlo simulations for a three-dimensional layered sample with finite geometry have been performed to confirm the detectability of Raman signals at all angles and at all sides of the object. We considered a non-transparent cuboid container (high density polyethylene) with explosive material (ammonium nitrate) inside. The simulation results were validated with experimental Raman intensities. Monte Carlo simulation results reveal that the ratio of sub-surface to surface signals improves at geometries other than backscattering. In addition, we demonstrate through simulations the effects of the absorption and scattering coefficients of the layers, and that of the diameter of the excitation beam. The advantage of collecting light from all possible 4 angles, over other collection modes, is that this technique is not geometry specific and molecular identification of layers underneath non-transparent surfaces can be obtained with minimal interference from the surface layer. To what extent all sides of the object will contribute to the total signal will depend on the absorption and scattering coefficients and the physical dimensions. Copyright (c) 2015 John Wiley & Sons, Ltd.

SeamSeg: Video Object Segmentation using Patch Seams

In this paper, we propose a technique for video object segmentation using patch seams across frames. Typically, seams, which are connected paths of low energy, are utilised for retargeting, where the primary aim is to reduce the image size while preserving the salient image contents. Here, we adapt the formulation of seams for temporal label propagation. The energy function associated with the proposed video seams provides temporal linking of patches across frames, to accurately segment the object. The proposed energy function takes into account the similarity of patches along the seam, temporal consistency of motion and spatial coherency of seams. Label propagation is achieved with high fidelity in the critical boundary regions, utilising the proposed patch seams. To achieve this without additional overheads, we curtail the error propagation by formulating boundary regions as rough-sets. The proposed approach out-perform state-of-the-art supervised and unsupervised algorithms, on benchmark datasets.

A Cause and Effect Analysis of Motion Trajectories for Modeling Actions

An action is typically composed of different parts of the object moving in particular sequences. The presence of different motions (represented as a 1D histogram) has been used in the traditional bag-of-words (BoW) approach for recognizing actions. However the interactions among the motions also form a crucial part of an action. Different object-parts have varying degrees of interactions with the other parts during an action cycle. It is these interactions we want to quantify in order to bring in additional information about the actions. In this paper we propose a causality based approach for quantifying the interactions to aid action classification. Granger causality is used to compute the cause and effect relationships for pairs of motion trajectories of a video. A 2D histogram descriptor for the video is constructed using these pairwise measures. Our proposed method of obtaining pairwise measures for videos is also applicable for large datasets. We have conducted experiments on challenging action recognition databases such as HMDB51 and UCF50 and shown that our causality descriptor helps in encoding additional information regarding the actions and performs on par with the state-of-the art approaches. Due to the complementary nature, a further increase in performance can be observed by combining our approach with state-of-the-art approaches.

A Bacterial Foraging Optimization and Learning Automata Based Feature Selection for Motor Imagery EEG Classification

Selection of relevant features is an open problem in Brain-computer interfacing (BCI) research. Sometimes, features extracted from brain signals are high dimensional which in turn affects the accuracy of the classifier. Selection of the most relevant features improves the performance of the classifier and reduces the computational cost of the system. In this study, we have used a combination of Bacterial Foraging Optimization and Learning Automata to determine the best subset of features from a given motor imagery electroencephalography (EEG) based BCI dataset. Here, we have employed Discrete Wavelet Transform to obtain a high dimensional feature set and classified it by Distance Likelihood Ratio Test. Our proposed feature selector produced an accuracy of 80.291% in 216 seconds.

Damage Detection Sensitivity, Specificity and Classification Data Analysis for SHM Systems Design, Verification and Validation

In this paper, we consider applying derived knowledge base regarding the sensitivity and specificity of damage(s) to be detected by an SHM system being designed and qualified. These efforts are necessary toward developing capabilities in SHM system to classify reliably various probable damages through sequence of monitoring, i.e., damage precursor identification, detection of damage and monitoring its progression. We consider the particular problem of visual and ultrasonic NDE based SHM system design requirements, where the damage detection sensitivity and specificity data definitions for a class of structural components are established. Methodologies for SHM system specification creation are discussed in details. Examples are shown to illustrate how the physics of damage detection scheme limits particular damage detection sensitivity and specificity and further how these information can be used in algorithms to combine various different NDE schemes in an SHM system to enhance efficiency and effectiveness. Statistical and data driven models to determine the sensitivity and probability of damage detection (POD) has been demonstrated for plate with varying one-sided line crack using optical and ultrasonic based inspection techniques.