Fast Learning VIEWNET Architectures for Recognizing 3-D Objects from Multiple 2-D Views

The recognition of 3-D objects from sequences of their 2-D views is modeled by a family of self-organizing neural architectures, called VIEWNET, that use View Information Encoded With NETworks. VIEWNET incorporates a preprocessor that generates a compressed but 2-D invariant representation of an image, a supervised incremental learning system that classifies the preprocessed representations into 2-D view categories whose outputs arc combined into 3-D invariant object categories, and a working memory that makes a 3-D object prediction by accumulating evidence from 3-D object category nodes as multiple 2-D views are experienced. The simplest VIEWNET achieves high recognition scores without the need to explicitly code the temporal order of 2-D views in working memory. Working memories are also discussed that save memory resources by implicitly coding temporal order in terms of the relative activity of 2-D view category nodes, rather than as explicit 2-D view transitions. Variants of the VIEWNET architecture may also be used for scene understanding by using a preprocessor and classifier that can determine both What objects are in a scene and Where they are located. The present VIEWNET preprocessor includes the CORT-X 2 filter, which discounts the illuminant, regularizes and completes figural boundaries, and suppresses image noise. This boundary segmentation is rendered invariant under 2-D translation, rotation, and dilation by use of a log-polar transform. The invariant spectra undergo Gaussian coarse coding to further reduce noise and 3-D foreshortening effects, and to increase generalization. These compressed codes are input into the classifier, a supervised learning system based on the fuzzy ARTMAP algorithm. Fuzzy ARTMAP learns 2-D view categories that are invariant under 2-D image translation, rotation, and dilation as well as 3-D image transformations that do not cause a predictive error. Evidence from sequence of 2-D view categories converges at 3-D object nodes that generate a response invariant under changes of 2-D view. These 3-D object nodes input to a working memory that accumulates evidence over time to improve object recognition. ln the simplest working memory, each occurrence (nonoccurrence) of a 2-D view category increases (decreases) the corresponding node's activity in working memory. The maximally active node is used to predict the 3-D object. Recognition is studied with noisy and clean image using slow and fast learning. Slow learning at the fuzzy ARTMAP map field is adapted to learn the conditional probability of the 3-D object given the selected 2-D view category. VIEWNET is demonstrated on an MIT Lincoln Laboratory database of l28x128 2-D views of aircraft with and without additive noise. A recognition rate of up to 90% is achieved with one 2-D view and of up to 98.5% correct with three 2-D views. The properties of 2-D view and 3-D object category nodes are compared with those of cells in monkey inferotemporal cortex.

Experimental investigation of thermal structures in regular three-dimensional falling films

Interfacial waves on the surface of a falling liquid film are known to modify heat and mass transfer. Under non-isothermal conditions, the wave topology is strongly influenced by the presence of thermocapillary (Marangoni) forces at the interface which leads to a destabilization of the film flow and potentially to critical film thinning. In this context, the present study investigates the evolution of the surface topology and the evolution of the surface temperature for the case of regularly excited solitary-type waves on a falling liquid film under the influence of a wall-side heat flux. Combining film thickness (chromatic confocal imaging) and surface temperature information (infrared thermography), interactions between hydrodynamics and thermocapillary forces are revealed. These include the formation of rivulets, film thinning and wave number doubling in spanwise direction. Distinct thermal structures on the films’ surface can be associated to characteristics of the surface topology.

Details of an integrated approach to three-dimensional dynamic fluid structure interaction

Fluid structure interaction, as applied to flexible structures, has wide application in diverse areas such as flutter in aircraft, wind response of buildings, flows in elastic pipes and blood vessels. Numerical modelling of dynamic fluid-structure interaction (DFSI) involves the coupling of fluid flow and structural mechanics, two fields that are conventionally modelled using two dissimilar methods, thus a single comprehensive computational model of both phenomena is a considerable challenge and until recently work in this area focused on one phenomenon and represented the behaviour of the other more simply. A single, finite volume unstructured mesh (FV-UM) spatial discretisation method has been employed on a single mesh for the entire domain. The Navier Stokes equations for fluid flow are solved using a SIMPLE type procedure and the Newmark b algorithm is employed for solving the dynamic equilibrium equations for linear elastic solid mechanics and mesh movement is achieved using a spring based mesh procedure for dynamic mesh movement. In the paper we describe a number of additional computation issues for the efficient and accurate modelling of three-dimensional, dynamic fluid-structure interaction problems.

Three-dimensional simulation of the influence of convection on dendritic solidification

Bulk and interdendritic flow during solidification alters the microstructure development, potentially leading to the formation of defects. In this paper, a 3D numerical model is presented for the simulation of dendritic growth in the presence of fluid flow in both liquid and semi-solid zones during solidification. The dendritic growth was solved by the combination of a stochastic nucleation approach with a finite difference solution of the solute diffusion equation and. a projection method solution of the Navier-Stokes equations. The technique was applied first to simulate the growth of a single dendrite in 2D and 3D in an isothermal environment with forced fluid flow. Significant differences were found in the evolution of dendritic morphology when comparing the 2D and 3D results. In 3D the upstream arm has a faster growth velocity due to easier flow around the perpendicular arms. This also promotes secondary arm formation on the upstream arm. The effect of fluid flow on columnar dendritic growth and micro-segregation in constrained solidification conditions is then simulated. For constrained growth, 2D simulations lead to even greater inaccuracies as compared to 3D.

Three-dimensional macroporous silicon photonic crystal with large photonic band gap

Three-dimensional photonic crystals based on macroporous silicon are fabricated by photoelectrochemical etching and subsequent focused-ion-beam drilling. Reflection measurements show a high reflection in the range of the stopgap and indicate the spectral position of the complete photonic band gap. The onset of diffraction which might influence the measurement is discussed.

Optimisation of radiotherapy for carcinoma of the parotid gland: a comparison of conventional, three-dimensional conformal, and intensity-modulated techniques

Background and purpose: To compare external beam radiotherapy techniques for parotid gland tumours using conventional radiotherapy (RT), three-dimensional conformal radiotherapy (3DCRT), and intensity-modulated radiotherapy (IMRT). To optimise the IMRT techniques, and to produce an IMRT class solution.Materials and methods: The planning target volume (PTV), contra-lateral parotid gland, oral cavity, brain-stem, brain and cochlea were outlined on CT planning scans of six patients with parotid gland tumours. Optimised conventional RT and 3DCRT plans were created and compared with inverse-planned IMRT dose distributions using dose-volume histograms. The aim was to reduce the radiation dose to organs at risk and improve the PTV dose distribution. A beam-direction optimisation algorithm was used to improve the dose distribution of the IMRT plans, and a class solution for parotid gland IMRT was investigated.Results: 3DCRT plans produced an equivalent PTV irradiation and reduced the dose to the cochlea, oral cavity, brain, and other normal tissues compared with conventional RT. IMRT further reduced the radiation dose to the cochlea and oral cavity compared with 3DCRT. For nine- and seven-field IMRT techniques, there was an increase in low-dose radiation to non-target tissue and the contra-lateral parotid gland. IMRT plans produced using three to five optimised intensity-modulated beam directions maintained the advantages of the more complex IMRT plans, and reduced the contra-lateral parotid gland dose to acceptable levels. Three- and four-field non-coplanar beam arrangements increased the volume of brain irradiated, and increased PTV dose inhomogeneity. A four-field class solution consisting of paired ipsilateral coplanar anterior and posterior oblique beams (15, 45, 145 and 170o from the anterior plane) was developed which maintained the benefits without the complexity of individual patient optimisation.Conclusions: For patients with parotid gland tumours, reduction in the radiation dose to critical normal tissues was demonstrated with 3DCRT compared with conventional RT. IMRT produced a further reduction in the dose to the cochlea and oral cavity. With nine and seven fields, the dose to the contra-lateral parotid gland was increased, but this was avoided by optimisation of the beam directions. The benefits of IMRT were maintained with three or four fields when the beam angles were optimised, but were also achieved using a four-field class solution. Clinical trials are required to confirm the clinical benefits of these improved dose distributions.

Mono- and dimethyl-1,2,4-triazolate as ligands for the mercuric ion in the cationic three-dimensional coordination polymer of [Hg-2(Mmt)(Dmt)(2)](NO3)(H2O))

Colourless crystals of [Hg-2(Mmt)(Dmt)(2)](NO3)(H2O) were obtained from a reaction of mercuric nitrate with nionomethyl- and dimethyl-1,2.4-triazolate (Mmt(-) and Dmt(-), respectively). In the crystal structure (monoclinic, C2/c (no. 15), a = 2579.4(4) b = 1231.1(2), c = 1634.8(2) pm, beta = 128.32(1)degrees V = 4073.3(11).10(6).pm(3): Z = 8, R-1 [I-0 > 2 sigma(I-0)]: 0.0355), half of the mercuric ions are essentially two-coordinate (Hg-N: 210-215 pm), the other half are tetrahedrally surrounded by N-donor atoms (Hg-N: 221, 225 pm) of the Mmt(-) and Dmt(-) anions. These three-N ligands construct a three-dimensional framework.

Three-Dimensional Analysis of Seepage below and around Hydraulic Structures

This paper investigates the problem of seepage under the floor of hydraulic structures considering the compartment of flow that seeps through the surrounding banks of the canal. A computer program, utilizing a finite-element method and capable of handling three-dimensional (3D) saturated–unsaturated flow problems, was used. Different ratios of canal width/differential head applied on the structure were studied. The results produced from the two-dimensional (2D) analysis were observed to deviate largely from that obtained from 3D analysis of the same problem, despite the fact that the porous medium was isotropic and homogeneous. For example, the exit gradient obtained from 3D analysis was as high as 2.5 times its value obtained from 2D analysis. Uplift force acting upwards on the structure has also increased by about 46% compared with its value obtained from the 2D solution. When the canal width/ differential head ratio was 10 or higher, the 3D results were comparable to the 2D results. It is recommended to construct a core of low permeability soil in the banks of canal to reduce the seepage losses, uplift force, and exit gradient.

On-line Emotion Recognition in a 3-D Activation-Valence-Time Continuum using Acoustic and Linguistic Cues

For many applications of emotion recognition, such as virtual agents, the system must select responses while the user is speaking. This requires reliable on-line recognition of the user’s affect. However most emotion recognition systems are based on turnwise processing. We present a novel approach to on-line emotion recognition from speech using Long Short-Term Memory Recurrent Neural Networks. Emotion is recognised frame-wise in a two-dimensional valence-activation continuum. In contrast to current state-of-the-art approaches, recognition is performed on low-level signal frames, similar to those used for speech recognition. No statistical functionals are applied to low-level feature contours. Framing at a higher level is therefore unnecessary and regression outputs can be produced in real-time for every low-level input frame. We also investigate the benefits of including linguistic features on the signal frame level obtained by a keyword spotter.