A coupled discriminative dictionary and transformation learning approach with applications to cross domain matching

Cross domain and cross-modal matching has many applications in the field of computer vision and pattern recognition. A few examples are heterogeneous face recognition, cross view action recognition, etc. This is a very challenging task since the data in two domains can differ significantly. In this work, we propose a coupled dictionary and transformation learning approach that models the relationship between the data in both domains. The approach learns a pair of transformation matrices that map the data in the two domains in such a manner that they share common sparse representations with respect to their own dictionaries in the transformed space. The dictionaries for the two domains are learnt in a coupled manner with an additional discriminative term to ensure improved recognition performance. The dictionaries and the transformation matrices are jointly updated in an iterative manner. The applicability of the proposed approach is illustrated by evaluating its performance on different challenging tasks: face recognition across pose, illumination and resolution, heterogeneous face recognition and cross view action recognition. Extensive experiments on five datasets namely, CMU-PIE, Multi-PIE, ChokePoint, HFB and IXMAS datasets and comparisons with several state-of-the-art approaches show the effectiveness of the proposed approach. (C) 2015 Elsevier B.V. All rights reserved.

A coupled discriminative dictionary and transformation learning approach with applications to cross domain matching

Cross domain and cross-modal matching has many applications in the field of computer vision and pattern recognition. A few examples are heterogeneous face recognition, cross view action recognition, etc. This is a very challenging task since the data in two domains can differ significantly. In this work, we propose a coupled dictionary and transformation learning approach that models the relationship between the data in both domains. The approach learns a pair of transformation matrices that map the data in the two domains in such a manner that they share common sparse representations with respect to their own dictionaries in the transformed space. The dictionaries for the two domains are learnt in a coupled manner with an additional discriminative term to ensure improved recognition performance. The dictionaries and the transformation matrices are jointly updated in an iterative manner. The applicability of the proposed approach is illustrated by evaluating its performance on different challenging tasks: face recognition across pose, illumination and resolution, heterogeneous face recognition and cross view action recognition. Extensive experiments on five datasets namely, CMU-PIE, Multi-PIE, ChokePoint, HFB and IXMAS datasets and comparisons with several state-of-the-art approaches show the effectiveness of the proposed approach. (C) 2015 Elsevier B.V. All rights reserved.

Toxicity and imaging of multi-walled carbon nanotubes in human macrophage cells.

Multi-walled carbon nanotubes (MWNTs) have been proposed for use in many applications and concerns about their potential effect on human health have led to the interest in understanding the interactions between MWNTs and human cells. One important technique is the visualisation of the intracellular distribution of MWNTs. We exposed human macrophage cells to unpurified MWNTs and found that a decrease in cell viability was correlated with uptake of MWNTs due to mainly necrosis. Cells treated with purified MWNTs and the main contaminant Fe(2)O(3) itself yielded toxicity only from the nanotubes and not from the Fe(2)O(3). We used 3-D dark-field scanning transmission electron microscopy (DF-STEM) tomography of freeze-dried whole cells as well as confocal and scanning electron microscopy (SEM) to image the cellular uptake and distribution of unpurified MWNTs. We observed that unpurified MWNTs entered the cell both actively and passively frequently inserting through the plasma membrane into the cytoplasm and the nucleus. These suggest that MWNTs may cause incomplete phagocytosis or mechanically pierce through the plasma membrane and result in oxidative stress and cell death.

Multi-channel signal separation

The separation of independent sources from mixed observed data is a fundamental and challenging problem. In many practical situations, observations may be modelled as linear mixtures of a number of source signals, i.e. a linear multi-input multi-output system. A typical example is speech recordings made in an acoustic environment in the presence of background noise and/or competing speakers. Other examples include EEG signals, passive sonar applications and cross-talk in data communications. In this paper, we propose iterative algorithms to solve the n × n linear time invariant system under two different constraints. Some existing solutions for 2 × 2 systems are reviewed and compared.

Multi-scale analysis of AFM tip and surface interactions

Thoroughly understanding AFM tip-surface interactions is crucial for many experimental studies and applications. It is important to realize that despite its simple appearance, the system of tip and sample surface involves multiscale interactions. In fact, the system is governed by a combination of molecular force (like the van der Waals force), its macroscopic representations (such as surface force) and gravitational force (a macroscopic force). Hence, in the system, various length scales are operative, from sub-nanoscale (at the molecular level) to the macroscopic scale. By integrating molecular forces into continuum equations, we performed a multiscale analysis and revealed the nonlocality effect between a tip and a rough solid surface and the mechanism governing liquid surface deformation and jumping. The results have several significant implications for practical applications. For instance, nonlocality may affect the measurement accuracy of surface morphology. At the critical state of liquid surface jump, the ratio of the gap between a tip and a liquid dome (delta) over the dome height (y(o)) is approximately (n-4) (for a large tip), which depends on the power law exponent n of the molecular interaction energy. These findings demonstrate that the multiscale analysis is not only useful but also necessary in the understanding of practical phenomena involving molecular forces. (c) 2007 Elsevier Ltd. All rights reserved.

An approximate likelihood method for estimating static parameters in multi-target tracking models

This paper considers a class of dynamic Spatial Point Processes (PP) that evolves over time in a Markovian fashion. This Markov in time PP is hidden and observed indirectly through another PP via thinning, displacement and noise. This statistical model is important for Multi object Tracking applications and we present an approximate likelihood based method for estimating the model parameters. The work is supported by an extensive numerical study.

Stimulation discomfort comparison of asynchronous and synchronous methods with multi-field electrodes

Functional Electrical Stimulation (FES) is a technique that consists on applying electrical current pulses to artificially activate motor nerve fibers and produce muscle contractions to achieve functional movements. The main applications of FES are within the rehabilitation field, in which this technique is used to aid recovery or to restore lost motor functions. People that benefit of FES are usually patients with neurological disorders which result in motor dysfunctions; most common patients include stroke and spinal cord injury (SCI). Neuroprosthesis are devices that have their basis in FES technique, and their aim is to bridge interrupted or damaged neural paths between the brain and upper or lower limbs. One of the aims of neuroprosthesis is to artificially generate muscle contractions that produce functional movements, and therefore, assist impaired people by making them able to perform activities of daily living (ADL). FES applies current pulses and stimulates nerve fibers by means of electrodes, which can be either implanted or surface electrodes. Both of them have advantages and disadvantages. Implanted electrodes need open surgery to place them next to the nerve root, so these electrodes carry many disadvantages that are produced by the use of invasive techniques. In return, as the electrodes are attached to the nerve, they make it easier to achieve selective functional movements. On the contrary, surface electrodes are not invasive and are easily attached or detached on the skin. Main disadvantages of surface electrodes are the difficulty of selectively stimulating nerve fibers and uncomfortable feeling perceived by users due to sensory nerves located in the skin. Electrical stimulation surface electrode technology has improved significantly through the years and recently, multi-field electrodes have been suggested. This multi-field or matrix electrode approach brings many advantages to FES; among them it is the possibility of easily applying different stimulation methods and techniques. The main goal of this thesis is therefore, to test two stimulation methods, which are asynchronous and synchronous stimulation, in the upper limb with multi-field electrodes. To this end, a purpose-built wrist torque measuring system and a graphic user interface were developed to measure wrist torque produced with each of the methods and to efficiently carry out the experiments. Then, both methods were tested on 15 healthy subjects and sensitivity results were analyzed for different cases. Results show that there are significant differences between methods regarding sensation in some cases, which can affect effectiveness or success of FES.