An adaptation framework for head-pose classification in dynamic multi-view scenarios

Multi-view head-pose estimation in low-resolution, dynamic scenes is difficult due to blurred facial appearance and perspective changes as targets move around freely in the environment. Under these conditions, acquiring sufficient training examples to learn the dynamic relationship between position, face appearance and head-pose can be very expensive. Instead, a transfer learning approach is proposed in this work. Upon learning a weighted-distance function from many examples where the target position is fixed, we adapt these weights to the scenario where target positions are varying. The adaptation framework incorporates reliability of the different face regions for pose estimation under positional variation, by transforming the target appearance to a canonical appearance corresponding to a reference scene location. Experimental results confirm effectiveness of the proposed approach, which outperforms state-of-the-art by 9.5% under relevant conditions. To aid further research on this topic, we also make DPOSE- a dynamic, multi-view head-pose dataset with ground-truth publicly available with this paper.

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.

Re-analysis of cryoEM data on HCV IRES bound to 40S subunit of human ribosome integrated with recent structural information suggests new contact regions between ribosomal proteins and HCV RNA

In this study, we combine available high resolution structural information on eukaryotic ribosomes with low resolution cryo-EM data on the Hepatitis C Viral RNA (IRES) human ribosome complex. Aided further by the prediction of RNA-protein interactions and restrained docking studies, we gain insights on their interaction at the residue level. We identified the components involved at the major and minor contact regions, and propose that there are energetically favorable local interactions between 40S ribosomal proteins and IRES domains. Domain II of the IRES interacts with ribosomal proteins S5 and S25 while the pseudoknot and the downstream domain IV region bind to ribosomal proteins S26, S28 and S5. We also provide support using UV cross-linking studies to validate our proposition of interaction between the S5 and IRES domains II and IV. We found that domain IIIe makes contact with the ribosomal protein S3a (S1e). Our model also suggests that the ribosomal protein S27 interacts with domain IIIc while S7 has a weak contact with a single base RNA bulge between junction IIIabc and IIId. The interacting residues are highly conserved among mammalian homologs while IRES RNA bases involved in contact do not show strict conservation. IRES RNA binding sites for S25 and S3a show the best conservation among related viral IRESs. The new contacts identified between ribosomal proteins and RNA are consistent with previous independent studies on RNA-binding properties of ribosomal proteins reported in literature, though information at the residue level is not available in previous studies.

Molecular Mechanism Underlying ATP-Induced Conformational Changes in the Nucleoprotein Filament of Mycobacterium smegmatis RecA

RecA plays a central role in bacterial DNA repair, homologous recombination, and restoration of stalled replication forks by virtue of its active extended nucleoprotein filament. Binding of ATP and its subsequent recognition by the carboxamide group of a highly conserved glutamine (GIn196 in MsRecA) have been implicated in the formation of active RecA nucleoprotein filaments. Although the mechanism of ATP-dependent structural transitions in RecA has been proposed on the basis of low-resolution electron microscopic reconstructions, the precise sequence of events that constitute these transitions is poorly understood. On the basis of biochemical and crystallographic analyses of MsRecA variants carrying mutations in highly conserved Gln196 and Arg198 residues, we propose that the disposition of the interprotomer interface is the structural basis of allosteric activation of RecA. Furthermore, this study accounts, for the contributions of several conserved amino acids to ATP hydrolysis and to the transition from collapsed to extended filament forms in Mycobacterium smegmatis RecA (MsRecA). In addition to their role in the inactive compressed state, the study reveals a role for GIn196 and Arg198 along with Phe219 in ATP hydrolysis in the active extended nucleoprotein filament. Finally, our data suggest that the primary, but not secondary, nucleotide binding site in MsRecA isomerizes into the ATP binding site present in the extended nucleoprotein filament.

Study of an object recognition algorithm

This project introduces an improvement of the vision capacity of the robot Robotino operating under ROS platform. A method for recognizing object class using binary features has been developed. The proposed method performs a binary classification of the descriptors of each training image to characterize the appearance of the object class. It presents the use of the binary descriptor based on the difference of gray intensity of the pixels in the image. It shows that binary features are suitable to represent object class in spite of the low resolution and the weak information concerning details of the object in the image. It also introduces the use of a boosting method (Adaboost) of feature selection al- lowing to eliminate redundancies and noise in order to improve the performance of the classifier. Finally, a kernel classifier SVM (Support Vector Machine) is trained with the available database and applied for predictions on new images. One possible future work is to establish a visual servo-control that is to say the reac- tion of the robot to the detection of the object.

Bright-field and fluorescence chip-scale microscopy for biological imaging

Optical microscopy is an essential tool in biological science and one of the gold standards for medical examinations. Miniaturization of microscopes can be a crucial stepping stone towards realizing compact, cost-effective and portable platforms for biomedical research and healthcare. This thesis reports on implementations of bright-field and fluorescence chip-scale microscopes for a variety of biological imaging applications. The term “chip-scale microscopy” refers to lensless imaging techniques realized in the form of mass-producible semiconductor devices, which transforms the fundamental design of optical microscopes.

Our strategy for chip-scale microscopy involves utilization of low-cost Complementary metal Oxide Semiconductor (CMOS) image sensors, computational image processing and micro-fabricated structural components. First, the sub-pixel resolving optofluidic microscope (SROFM), will be presented, which combines microfluidics and pixel super-resolution image reconstruction to perform high-throughput imaging of fluidic samples, such as blood cells. We discuss design parameters and construction of the device, as well as the resulting images and the resolution of the device, which was 0.66 µm at the highest acuity. The potential applications of SROFM for clinical diagnosis of malaria in the resource-limited settings is discussed.

Next, the implementations of ePetri, a self-imaging Petri dish platform with microscopy resolution, are presented. Here, we simply place the sample of interest on the surface of the image sensor and capture the direct shadow images under the illumination. By taking advantage of the inherent motion of the microorganisms, we achieve high resolution (~1 µm) imaging and long term culture of motile microorganisms over ultra large field-of-view (5.7 mm × 4.4 mm) in a specialized ePetri platform. We apply the pixel super-resolution reconstruction to a set of low-resolution shadow images of the microorganisms as they move across the sensing area of an image sensor chip and render an improved resolution image. We perform longitudinal study of Euglena gracilis cultured in an ePetri platform and image based analysis on the motion and morphology of the cells. The ePetri device for imaging non-motile cells are also demonstrated, by using the sweeping illumination of a light emitting diode (LED) matrix for pixel super-resolution reconstruction of sub-pixel shifted shadow images. Using this prototype device, we demonstrate the detection of waterborne parasites for the effective diagnosis of enteric parasite infection in resource-limited settings.

Then, we demonstrate the adaptation of a smartphone’s camera to function as a compact lensless microscope, which uses ambient illumination as its light source and does not require the incorporation of a dedicated light source. The method is also based on the image reconstruction with sweeping illumination technique, where the sequence of images are captured while the user is manually tilting the device around any ambient light source, such as the sun or a lamp. Image acquisition and reconstruction is performed on the device using a custom-built android application, constructing a stand-alone imaging device for field applications. We discuss the construction of the device using a commercial smartphone and demonstrate the imaging capabilities of our system.

Finally, we report on the implementation of fluorescence chip-scale microscope, based on a silo-filter structure fabricated on the pixel array of a CMOS image sensor. The extruded pixel design with metal walls between neighboring pixels successfully guides fluorescence emission through the thick absorptive filter to the photodiode layer of a pixel. Our silo-filter CMOS image sensor prototype achieves 13-µm resolution for fluorescence imaging over a wide field-of-view (4.8 mm × 4.4 mm). Here, we demonstrate bright-field and fluorescence longitudinal imaging of living cells in a compact, low-cost configuration.

Mechanistic aspects of the Ziegler-Natta polymerization

The isotope effect on propagation rate was determined for four homogeneous ethylene polymerization systems. The catalytic system Cp_2Ti(Et)Cl + EtA1Cl_2 has a k^H_p/k^D_p = 1.035 ± 0.03. This result strongly supports an insertion mechanism which does not involve a hydrogen migration during the rate determining step of propagation (Cossee mechanism). Three metal-alkyl free systems were also studied. The catalyst I_2 (PMe_3)_3Ta(neopentylidene)(H) has a k^H_p/k^D_p = 1.709. It is interpreted as a primary isotope effect involving a non-linear a-hydrogen migration during the rate determining step of propagation (Green mechanism). The lanthanide complexes Cp*_2LuMe•Et_2O and Cp*_2YbMe•Et_2O have a k^H_p/k^D_p = 1.46 and 1.25, respectively. They are interpreted as primary isotope effects due to a partial hydrogen migration during the rate determining step of propagation.

The presence of a precoordination or other intermediate species during the polymerization of ethylene by the mentioned metal-alkyl free catalysts was sought by low temperature NMR spectroscopy. However, no evidence for such species was found. If they exist, their concentrations are very small or their lifetimes are shorter than the NMR time scale.

Two titanocene (alkenyl)chlorides (hexenyl 1 and heptenyl 2 were prepared from titanocene dichloride and a THF solution of the corresponding alkenylmagnesium chloride. They do not cyclize in solution when alone, but cyclization to their respective titanocene(methyl(cycloalkyl) chlorides occurs readily in the presence of a Lewis acid. It is demonstrated that such cyclization occurs with the alkenyl ligand within the coordination sphere of the titanium atom. Cyclization of 1 with EtAlCl_2 at 0°C occurs in less than 95 msec (ethylene insertion time), as shown by the presence of 97% cyclopentyl-capped oligomers when polymerizing ethylene with this system. Some alkyl exchange occurs (3%). Cyclization of 2 is slower under the same reaction conditions and is not complete in 95 msec as shown by the presence of both cyclohexyl-capped oligomers (35%) and odd number α-olefin oligomers (50%). Alkyl exchange is more extensive as evidenced by the even number n-alkanes (15%).

Cyclization of 2-d_1 (titanocene(hept-6-en-1-yl-1-d_1)chloride) with EtA1Cl_2 demonstrated that for this system there is no α-hydrogen participation during said process. The cyclization is believed to occur by a Cossee-type mechanism. There was no evidence for precoordination of the alkenyl double bond during the cyclization process.

Neocarzinostatin chromophore: structural, mechanistic, and synthetic studies

Neocarzinostatin chromophore 1 is the active component of the antitumor antibiotic neocarzinostatin (NCS). The chromophore reacts with thiols to form a highly strained cumulene-enyne species which rapidly rearranges to a biradical intermediate which can abstract hydrogen atoms from DNA, leading to strand cleavage. DNA damage is the proposed source of biological activity for NCS. The structure of the methyl thioglycolate monoadduct 2 of NCS chromophore, including the absolute stereochemistry, was determined by NMR studies. The presence of the cumulene-enyne intermediate and the rearrangement to a biradical were supported by data from low temperature NMR investigations. Also included are synthetic approaches to NCS chromophore model compounds based on intramolecular addition of an acetylide to an aldehyde.

Structural Characterizations of the Dimeric Anti-HIV Antibody 2G12 and the HIV-2 Envelope Glycoprotein

More than thirty years after the discovery that Human Immunodeficiency Virus (HIV) was the causative agent of Acquired Immunodeficiency Syndrome (AIDS), the disease remains pandemic as long as no effective universal vaccine is found. Over 34 million individuals in the world are infected with the virus, and the vast majority of them have no access to the antiretroviral therapies that have largely reduced HIV to a chronic disease in the developed world. The first chapter of this thesis introduces the history of the virus. The key to the infectious mechanism of the virus lies in its envelope glycoprotein (Env), a trimeric spike on the viral surface that utilizes host T cell receptors for entry. Though HIV-1 Env is immunogenic, most infected patients do not mount an effective neutralizing antibody response against it. Broadly-neutralizing anti-Env antibodies (bNAbs) present in the serum of a minority of infected individuals are usually sufficient to prevent the progression to full blown AIDS. Thus, the molecular details of these bNAbs as well as the antibody-antigen interface are of prime interest for structural studies, as insight gained would contribute to the design of a more effective immunogen and potential vaccine candidate. The second chapter of this thesis describes the low-resolution crystal structure of one such antibody, 2G12 dimer, which targets a high mannose epitope on the surface of Env. Patients infected with HIV-2, a related virus with ~35% sequence identity in the Env region, can generally mount a robust antibody response sufficient for viral control for reasons still unknown. The final two chapters of this thesis focus on the first reported structural studies of HIV-2 Env, the molecular details of which may inform HIV-1 therapy and immunogen design.

Ab initio study of anharmonicity in high pressure Cmca-4 metallic hydrogen

Hydrogen is the only atom for which the Schr odinger equation is solvable. Consisting only of a proton and an electron, hydrogen is the lightest element and, nevertheless, is far from being simple. Under ambient conditions, it forms diatomic molecules H2 in gas phase, but di erent temperature and pressures lead to a complex phase diagram, which is not completely known yet. Solid hydrogen was rst documented in 1899 [1] and was found to be isolating. At higher pressures, however, hydrogen can be metallized. In 1935 Wigner and Huntington predicted that the metallization pressure would be 25 GPa [2], where molecules would disociate to form a monoatomic metal, as alkali metals that lie below hydrogen in the periodic table. The prediction of the metallization pressure turned out to be wrong: metallic hydrogen has not been found yet, even under a pressure as high as 320 GPa. Nevertheless, extrapolations based on optical measurements suggest that a metallic phase may be attained at 450 GPa [3]. The interest of material scientist in metallic hydrogen can be attributed, at least to a great extent, to Ashcroft, who in 1968 suggested that such a system could be a hightemperature superconductor [4]. The temperature at which this material would exhibit a transition from a superconducting to a non-superconducting state (Tc) was estimated to be around room temperature. The implications of such a statement are very interesting in the eld of astrophysics: in planets that contain a big quantity of hydrogen and whose temperature is below Tc, superconducting hydrogen may be found, specially at the center, where the gravitational pressure is high. This might be the case of Jupiter, whose proportion of hydrogen is about 90%. There are also speculations suggesting that the high magnetic eld of Jupiter is due to persistent currents related to the superconducting phase [5]. Metallization and superconductivity of hydrogen has puzzled scientists for decades, and the community is trying to answer several questions. For instance, what is the structure of hydrogen at very high pressures? Or a more general one: what is the maximum Tc a phonon-mediated superconductor can have [6]? A great experimental e ort has been carried out pursuing metallic hydrogen and trying to answer the questions above; however, the characterization of solid phases of hydrogen is a hard task. Achieving the high pressures needed to get the sought phases requires advanced technologies. Diamond anvil cells (DAC) are commonly used devices. These devices consist of two diamonds with a tip of small area; for this reason, when a force is applied, the pressure exerted is very big. This pressure is uniaxial, but it can be turned into hydrostatic pressure using transmitting media. Nowadays, this method makes it possible to reach pressures higher than 300 GPa, but even at this pressure hydrogen does not show metallic properties. A recently developed technique that is an improvement of DAC can reach pressures as high as 600 GPa [7], so it is a promising step forward in high pressure physics. Another drawback is that the electronic density of the structures is so low that X-ray di raction patterns have low resolution. For these reasons, ab initio studies are an important source of knowledge in this eld, within their limitations. When treating hydrogen, there are many subtleties in the calculations: as the atoms are so light, the ions forming the crystalline lattice have signi cant displacements even when temperatures are very low, and even at T=0 K, due to Heisenberg's uncertainty principle. Thus, the energy corresponding to this zero-point (ZP) motion is signi cant and has to be included in an accurate determination of the most stable phase. This has been done including ZP vibrational energies within the harmonic approximation for a range of pressures and at T=0 K, giving rise to a series of structures that are stable in their respective pressure ranges [8]. Very recently, a treatment of the phases of hydrogen that includes anharmonicity in ZP energies has suggested that relative stability of the phases may change with respect to the calculations within the harmonic approximation [9]. Many of the proposed structures for solid hydrogen have been investigated. Particularly, the Cmca-4 structure, which was found to be the stable one from 385-490 GPa [8], is metallic. Calculations for this structure, within the harmonic approximation for the ionic motion, predict a Tc up to 242 K at 450 GPa [10]. Nonetheless, due to the big ionic displacements, the harmonic approximation may not su ce to describe correctly the system. The aim of this work is to apply a recently developed method to treat anharmonicity, the stochastic self-consistent harmonic approximation (SSCHA) [11], to Cmca-4 metallic hydrogen. This way, we will be able to study the e ects of anharmonicity in the phonon spectrum and to try to understand the changes it may provoque in the value of Tc. The work is structured as follows. First we present the theoretical basis of the calculations: Density Functional Theory (DFT) for the electronic calculations, phonons in the harmonic approximation and the SSCHA. Then we apply these methods to Cmca-4 hydrogen and we discuss the results obtained. In the last chapter we draw some conclusions and propose possible future work.

Deep near-infrared spectroscopy of high-redshift galaxies: the physical growth of passive systems

The assembly history of massive galaxies is one of the most important aspects of galaxy formation and evolution. Although we have a broad idea of what physical processes govern the early phases of galaxy evolution, there are still many open questions. In this thesis I demonstrate the crucial role that spectroscopy can play in a physical understanding of galaxy evolution. I present deep near-infrared spectroscopy for a sample of high-redshift galaxies, from which I derive important physical properties and their evolution with cosmic time. I take advantage of the recent arrival of efficient near-infrared detectors to target the rest-frame optical spectra of z > 1 galaxies, from which many physical quantities can be derived. After illustrating the applications of near-infrared deep spectroscopy with a study of star-forming galaxies, I focus on the evolution of massive quiescent systems.

Most of this thesis is based on two samples collected at the W. M. Keck Observatory that represent a significant step forward in the spectroscopic study of z > 1 quiescent galaxies. All previous spectroscopic samples at this redshift were either limited to a few objects, or much shallower in terms of depth. Our first sample is composed of 56 quiescent galaxies at 1 < z < 1.6 collected using the upgraded red arm of the Low Resolution Imaging Spectrometer (LRIS). The second consists of 24 deep spectra of 1.5 < z < 2.5 quiescent objects observed with the Multi-Object Spectrometer For Infra-Red Exploration (MOSFIRE). Together, these spectra span the critical epoch 1 < z < 2.5, where most of the red sequence is formed, and where the sizes of quiescent systems are observed to increase significantly.

We measure stellar velocity dispersions and dynamical masses for the largest number of z > 1 quiescent galaxies to date. By assuming that the velocity dispersion of a massive galaxy does not change throughout its lifetime, as suggested by theoretical studies, we match galaxies in the local universe with their high-redshift progenitors. This allows us to derive the physical growth in mass and size experienced by individual systems, which represents a substantial advance over photometric inferences based on the overall galaxy population. We find a significant physical growth among quiescent galaxies over 0 < z < 2.5 and, by comparing the slope of growth in the mass-size plane dlogR_e/dlogM_∗ with the results of numerical simulations, we can constrain the physical process responsible for the evolution. Our results show that the slope of growth becomes steeper at higher redshifts, yet is broadly consistent with minor mergers being the main process by which individual objects evolve in mass and size.

By fitting stellar population models to the observed spectroscopy and photometry we derive reliable ages and other stellar population properties. We show that the addition of the spectroscopic data helps break the degeneracy between age and dust extinction, and yields significantly more robust results compared to fitting models to the photometry alone. We detect a clear relation between size and age, where larger galaxies are younger. Therefore, over time the average size of the quiescent population will increase because of the contribution of large galaxies recently arrived to the red sequence. This effect, called progenitor bias, is different from the physical size growth discussed above, but represents another contribution to the observed difference between the typical sizes of low- and high-redshift quiescent galaxies. By reconstructing the evolution of the red sequence starting at z ∼ 1.25 and using our stellar population histories to infer the past behavior to z ∼ 2, we demonstrate that progenitor bias accounts for only half of the observed growth of the population. The remaining size evolution must be due to physical growth of individual systems, in agreement with our dynamical study.

Finally, we use the stellar population properties to explore the earliest periods which led to the formation of massive quiescent galaxies. We find tentative evidence for two channels of star formation quenching, which suggests the existence of two independent physical mechanisms. We also detect a mass downsizing, where more massive galaxies form at higher redshift, and then evolve passively. By analyzing in depth the star formation history of the brightest object at z > 2 in our sample, we are able to put constraints on the quenching timescale and on the properties of its progenitor.

A consistent picture emerges from our analyses: massive galaxies form at very early epochs, are quenched on short timescales, and then evolve passively. The evolution is passive in the sense that no new stars are formed, but significant mass and size growth is achieved by accreting smaller, gas-poor systems. At the same time the population of quiescent galaxies grows in number due to the quenching of larger star-forming galaxies. This picture is in agreement with other observational studies, such as measurements of the merger rate and analyses of galaxy evolution at fixed number density.

I. The 3.7 Å crystal structure of horse heart ferricytochrome C. II. The application of the Karle-Hauptman tangent formula to protein phasing

I. The 3.7 Å Crystal Structure of Horse Heart Ferricytochrome C.

The crystal structure of horse heart ferricytochrome c has been determined to a resolution of 3.7 Å using the multiple isomorphous replacement technique. Two isomorphous derivatives were used in the analysis, leading to a map with a mean figure of merit of 0.458. The quality of the resulting map was extremely high, even though the derivative data did not appear to be of high quality.

Although it was impossible to fit the known amino acid sequence to the calculated structure in an unambiguous way, many important features of the molecule could still be determined from the 3.7 Å electron density map. Among these was the fact that cytochrome c contains little or no α-helix. The polypeptide chain appears to be wound about the heme group in such a way as to form a loosely packed hydrophobic core in the molecule.

The heme group is located in a cleft on the molecule with one edge exposed to the solvent. The fifth coordinating ligand is His 18 and the sixth coordinating ligand is probably neither His 26 nor His 33.

The high resolution analysis of cytochrome c is now in progress and should be completed within the next year.

II. The Application of the Karle-Hauptman Tangent Formula to Protein Phasing.

The Karle-Hauptman tangent formula has been shown to be applicable to the refinement of previously determined protein phases. Tests were made with both the cytochrome c data from Part I and a theoretical structure based on the myoglobin molecule. The refinement process was found to be highly dependent upon the manner in which the tangent formula was applied. Iterative procedures did not work well, at least at low resolution.

The tangent formula worked very well in selecting the true phase from the two possible phase choices resulting from a single isomorphous replacement phase analysis. The only restriction on this application is that the heavy atoms form a non-centric cluster in the unit cell.

Pages 156 through 284 in this Thesis consist of previously published papers relating to the above two sections. References to these papers can be found on page 155.

Study of crystal structure and phase transition studies in perovskite-type oxides using powder-diffraction techniques and symmetry-mode analysis.

251 p.

Short-channel polymer field-effect-transistor fabrication using spin-coating-induced edge template and ink-jet printing

A method to fabricate polymer field-effect transistors with submicron channel lengths is described. A thin polymer film is spin coated on a prepatterned resist with a low resolution to create a thickness contrast in the overcoated polymer layer. After plasma and solvent etching, a submicron-sized line structure, which templates the contour of the prepattern, is obtained. A further lift-off process is applied to define source-drain electrodes of transistors. With a combination of ink-jet printing, transistors with channel length down to 400 nm have been fabricated by this method. We show that drive current density increases as expected, while the on/off current ratio 106 is achieved. © 2005 American Institute of Physics.

Face recognition from video

In spite of over two decades of intense research, illumination and pose invariance remain prohibitively challenging aspects of face recognition for most practical applications. The objective of this work is to recognize faces using video sequences both for training and recognition input, in a realistic, unconstrained setup in which lighting, pose and user motion pattern have a wide variability and face images are of low resolution. In particular there are three areas of novelty: (i) we show how a photometric model of image formation can be combined with a statistical model of generic face appearance variation, learnt offline, to generalize in the presence of extreme illumination changes; (ii) we use the smoothness of geodesically local appearance manifold structure and a robust same-identity likelihood to achieve invariance to unseen head poses; and (iii) we introduce an accurate video sequence "reillumination" algorithm to achieve robustness to face motion patterns in video. We describe a fully automatic recognition system based on the proposed method and an extensive evaluation on 171 individuals and over 1300 video sequences with extreme illumination, pose and head motion variation. On this challenging data set our system consistently demonstrated a nearly perfect recognition rate (over 99.7%), significantly outperforming state-of-the-art commercial software and methods from the literature. © Springer-Verlag Berlin Heidelberg 2006.