Heteroscedastic probabilistic linear discriminant analysis for manifold learning in video-based face recognition

To recognize faces in video, face appearances have been widely modeled as piece-wise local linear models which linearly approximate the smooth yet non-linear low dimensional face appearance manifolds. The choice of representations of the local models is crucial. Most of the existing methods learn each local model individually meaning that they only anticipate variations within each class. In this work, we propose to represent local models as Gaussian distributions which are learned simultaneously using the heteroscedastic probabilistic linear discriminant analysis (PLDA). Each gallery video is therefore represented as a collection of such distributions. With the PLDA, not only the within-class variations are estimated during the training, the separability between classes is also maximized leading to an improved discrimination. The heteroscedastic PLDA itself is adapted from the standard PLDA to approximate face appearance manifolds more accurately. Instead of assuming a single global within-class covariance, the heteroscedastic PLDA learns different within-class covariances specific to each local model. In the recognition phase, a probe video is matched against gallery samples through the fusion of point-to-model distances. Experiments on the Honda and MoBo datasets have shown the merit of the proposed method which achieves better performance than the state-of-the-art technique.

Breast cancer detection from thermal images using bispectral invariant features

Highly sensitive infrared (IR) cameras provide high-resolution diagnostic images of the temperature and vascular changes of breasts. These images can be processed to emphasize hot spots that exhibit early and subtle changes owing to pathology. The resulting images show clusters that appear random in shape and spatial distribution but carry class dependent information in shape and texture. Automated pattern recognition techniques are challenged because of changes in location, size and orientation of these clusters. Higher order spectral invariant features provide robustness to such transformations and are suited for texture and shape dependent information extraction from noisy images. In this work, the effectiveness of bispectral invariant features in diagnostic classification of breast thermal images into malignant, benign and normal classes is evaluated and a phase-only variant of these features is proposed. High resolution IR images of breasts, captured with measuring accuracy of ±0.4% (full scale) and temperature resolution of 0.1 °C black body, depicting malignant, benign and normal pathologies are used in this study. Breast images are registered using their lower boundaries, automatically extracted using landmark points whose locations are learned during training. Boundaries are extracted using Canny edge detection and elimination of inner edges. Breast images are then segmented using fuzzy c-means clustering and the hottest regions are selected for feature extraction. Bispectral invariant features are extracted from Radon projections of these images. An Adaboost classifier is used to select and fuse the best features during training and then classify unseen test images into malignant, benign and normal classes. A data set comprising 9 malignant, 12 benign and 11 normal cases is used for evaluation of performance. Malignant cases are detected with 95% accuracy. A variant of the features using the normalized bispectrum, which discards all magnitude information, is shown to perform better for classification between benign and normal cases, with 83% accuracy compared to 66% for the original.

Improving opportunity recognition and business performance in small and medium manufacturing enterprises through design innovation programs

Increasing awareness of the benefits of stimulating entrepreneurial behaviour in small and medium enterprises has fostered strong interest in innovation programs. Recently many western countries have invested in design innovation for better firm performance. This research presents some early findings from a study of companies that participated in a holistic approach to design innovation, where the outcomes include better business performance and better market positioning in global markets. Preliminary findings from in-depth semi-structured interviews indicate the importance of firm openness to new ways of working and to developing new processes of strategic entrepreneurship. Implications for theory and practice are discussed.

How the environment affects opportunity recognition by entrepreneurs at the bottom of the pyramid

There is an army of bottom of the pyramid entrepreneurs (BOPE) who have the potential to transform developing economies, if they can identify and exploit business opportunities. BOPE could have unidentified resources that could lead to the recognition of radical new opportunities. This study paper asks how environmental factors and identification of resources affect Opportunity Recognition by BOP entrepreneurs in developing economies. To investigate this research question we conduct a literature review and plan semi-structured interviews of existing and nascent entrepreneurs in the largest and arguably the poorest country in Africa, the Democratic Republic of the Congo. In this paper we review the context of BOPE and describe the methodology we will use to gather and analyse data. Finally, we describe our access to suitable respondents for this study and how it will be conducted.

I-vector based speaker recognition using advanced channel compensation techniques

This paper investigates advanced channel compensation techniques for the purpose of improving i-vector speaker verification performance in the presence of high intersession variability using the NIST 2008 and 2010 SRE corpora. The performance of four channel compensation techniques: (a) weighted maximum margin criterion (WMMC), (b) source-normalized WMMC (SN-WMMC), (c) weighted linear discriminant analysis (WLDA), and; (d) source-normalized WLDA (SN-WLDA) have been investigated. We show that, by extracting the discriminatory information between pairs of speakers as well as capturing the source variation information in the development i-vector space, the SN-WLDA based cosine similarity scoring (CSS) i-vector system is shown to provide over 20% improvement in EER for NIST 2008 interview and microphone verification and over 10% improvement in EER for NIST 2008 telephone verification, when compared to SN-LDA based CSS i-vector system. Further, score-level fusion techniques are analyzed to combine the best channel compensation approaches, to provide over 8% improvement in DCF over the best single approach, (SN-WLDA), for NIST 2008 interview/ telephone enrolment-verification condition. Finally, we demonstrate that the improvements found in the context of CSS also generalize to state-of-the-art GPLDA with up to 14% relative improvement in EER for NIST SRE 2010 interview and microphone verification and over 7% relative improvement in EER for NIST SRE 2010 telephone verification.

Testing the accuracy of luminance maps acquired by smart phone cameras

This paper looks at the accuracy of using the built-in camera of smart phones and free software as an economical way to quantify and analyse light exposure by producing luminance maps from High Dynamic Range (HDR) images. HDR images were captured with an Apple iPhone 4S to capture a wide variation of luminance within an indoor and outdoor scene. The HDR images were then processed using Photosphere software (Ward, 2010.) to produce luminance maps, where individual pixel values were compared with calibrated luminance meter readings. This comparison has shown an average luminance error of ~8% between the HDR image pixel values and luminance meter readings, when the range of luminances in the image is limited to approximately 1,500cd/m2.

The effect of project information on the accuracy of building price forecasts

A fundamental proposition is that the accuracy of the designer's tender price forecasts is positively correlated with the amount of information available for that project. The paper describes an empirical study of the effects of the quantity of information available on practicing Quantity Surveyors' forecasting accuracy. The methodology involved the surveyors repeatedly revising tender price forecasts on receipt of chunks of project information. Each of twelve surveyors undertook two projects and selected information chunks from a total of sixteen information types. The analysis indicated marked differences in accuracy between different project types and experts/non-experts. The expert surveyors' forecasts were not found to be significantly improved by information other than that of basic building type and size, even after eliminating project type effects. The expert surveyors' forecasts based on the knowledge of building type and size alone were, however, found to be of similar accuracy to that of average practitioners pricing full bills of quantities.

Factors affecting the accuracy of engineers' estimates

Emerging sciences, such as conceptual cost estimating, seem to have to go through two phases. The first phase involves reducing the field of study down to its basic ingredients - from systems development to technological development (techniques) to theoretical development. The second phase operates in the direction in building up techniques from theories, and systems from techniques. Cost estimating is clearly and distinctly still in the first phase. A great deal of effort has been put into the development of both manual and computer based cost estimating systems during this first phase and, to a lesser extent, the development of a range of techniques that can be used (see, for instance, Ashworth & Skitmore, 1986). Theoretical developments have not, as yet, been forthcoming. All theories need the support of some observational data and cost estimating is not likely to be an exception. These data do not need to be complete in order to build theories. As it is possible to construct an image of a prehistoric animal such as the brontosaurus from only a few key bones and relics, so a theory of cost estimating may possibly be found on a few factual details. The eternal argument of empiricists and deductionists is that, as theories need factual support, so do we need theories in order to know what facts to collect. In cost estimating, the basic facts of interest concern accuracy, the cost of achieving this accuracy, and the trade off between the two. When cost estimating theories do begin to emerge, it is highly likely that these relationships will be central features. This paper presents some of the facts we have been able to acquire regarding one part of this relationship - accuracy, and its influencing factors. Although some of these factors, such as the amount of information used in preparing the estimate, will have cost consequences, we have not yet reached the stage of quantifying these costs. Indeed, as will be seen, many of the factors do not involve any substantial cost considerations. The absence of any theory is reflected in the arbitrary manner in which the factors are presented. Rather, the emphasis here is on the consideration of purely empirical data concerning estimating accuracy. The essence of good empirical research is to .minimize the role of the researcher in interpreting the results of the study. Whilst space does not allow a full treatment of the material in this manner, the principle has been adopted as closely as possible to present results in an uncleaned and unbiased way. In most cases the evidence speaks for itself. The first part of the paper reviews most of the empirical evidence that we have located to date. Knowledge of any work done, but omitted here would be most welcome. The second part of the paper presents an analysis of some recently acquired data pertaining to this growing subject.

Accuracy in estimating

Several methods of estimating the costs or price of construction projects are now available for use in the construction industry. It is difficult due to the conservative approach of estimators and quantity surveyors, and the fact that the industry is undergoing one of its deepest recessions this century, to implement any changes in these processes. Several methods have been tried and tested and probably discarded forever, whereas other methods are still in their infancy. There is also a movement towards greater use of the computer, whichever method seems to be adopted. An important consideration with any method of estimating is the accuracy by which costs can be calculated. Any improvement in this consideration will be welcomed by a11 parties, because existing methods are poor when measured by this criteria. Estimating, particularly by contractors, has always carried some mystic, and many of the processes discussed both in the classroom and in practice are little more than fallacy when properly investigated. What makes an estimator or quantity surveyor good at forecasting the right price? To what extent does human behaviour influence or have a part to play? These and some of the other aspects of effective estimating are now examined in more detail.

A practical comparison of manual and autonomous methods for acoustic monitoring

1. Autonomous acoustic recorders are widely available and can provide a highly efficient method of species monitoring, especially when coupled with software to automate data processing. However, the adoption of these techniques is restricted by a lack of direct comparisons with existing manual field surveys. 2. We assessed the performance of autonomous methods by comparing manual and automated examination of acoustic recordings with a field-listening survey, using commercially available autonomous recorders and custom call detection and classification software. We compared the detection capability, time requirements, areal coverage and weather condition bias of these three methods using an established call monitoring programme for a nocturnal bird, the little spotted kiwi(Apteryx owenii). 3. The autonomous recorder methods had very high precision (>98%) and required <3% of the time needed for the field survey. They were less sensitive, with visual spectrogram inspection recovering 80% of the total calls detected and automated call detection 40%, although this recall increased with signal strength. The areal coverage of the spectrogram inspection and automatic detection methods were 85% and 42% of the field survey. The methods using autonomous recorders were more adversely affected by wind and did not show a positive association between ground moisture and call rates that was apparent from the field counts. However, all methods produced the same results for the most important conservation information from the survey: the annual change in calling activity. 4. Autonomous monitoring techniques incur different biases to manual surveys and so can yield different ecological conclusions if sampling is not adjusted accordingly. Nevertheless, the sensitivity, robustness and high accuracy of automated acoustic methods demonstrate that they offer a suitable and extremely efficient alternative to field observer point counts for species monitoring.

Insect odour perception: recognition of odour components by flower foraging moths

Odours emitted by flowers are complex blends of volatile compounds. These odours are learnt by flower-visiting insect species, improving their recognition of rewarding flowers and thus foraging efficiency. We investigated the flexibility of floral odour learning by testing whether adult moths recognize single compounds common to flowers on which they forage. Dual choice preference tests on Helicoverpa armigera moths allowed free flying moths to forage on one of three flower species; Argyranthemum frutescens (federation daisy), Cajanus cajan (pigeonpea) or Nicotiana tabacum (tobacco). Results showed that, (i) a benzenoid (phenylacetaldehyde) and a monoterpene (linalool) were subsequently recognized after visits to flowers that emitted these volatile constituents, (ii) in a preference test, other monoterpenes in the flowers' odour did not affect the moths' ability to recognize the monoterpene linalool and (iii) relative preferences for two volatiles changed after foraging experience on a single flower species that emitted both volatiles. The importance of using free flying insects and real flowers to understand the mechanisms involved in floral odour learning in nature are discussed in the context of our findings.

Robust image hash functions using higher order spectra

Robust hashing is an emerging field that can be used to hash certain data types in applications unsuitable for traditional cryptographic hashing methods. Traditional hashing functions have been used extensively for data/message integrity, data/message authentication, efficient file identification and password verification. These applications are possible because the hashing process is compressive, allowing for efficient comparisons in the hash domain but non-invertible meaning hashes can be used without revealing the original data. These techniques were developed with deterministic (non-changing) inputs such as files and passwords. For such data types a 1-bit or one character change can be significant, as a result the hashing process is sensitive to any change in the input. Unfortunately, there are certain applications where input data are not perfectly deterministic and minor changes cannot be avoided. Digital images and biometric features are two types of data where such changes exist but do not alter the meaning or appearance of the input. For such data types cryptographic hash functions cannot be usefully applied. In light of this, robust hashing has been developed as an alternative to cryptographic hashing and is designed to be robust to minor changes in the input. Although similar in name, robust hashing is fundamentally different from cryptographic hashing. Current robust hashing techniques are not based on cryptographic methods, but instead on pattern recognition techniques. Modern robust hashing algorithms consist of feature extraction followed by a randomization stage that introduces non-invertibility and compression, followed by quantization and binary encoding to produce a binary hash output. In order to preserve robustness of the extracted features, most randomization methods are linear and this is detrimental to the security aspects required of hash functions. Furthermore, the quantization and encoding stages used to binarize real-valued features requires the learning of appropriate quantization thresholds. How these thresholds are learnt has an important effect on hashing accuracy and the mere presence of such thresholds are a source of information leakage that can reduce hashing security. This dissertation outlines a systematic investigation of the quantization and encoding stages of robust hash functions. While existing literature has focused on the importance of quantization scheme, this research is the first to emphasise the importance of the quantizer training on both hashing accuracy and hashing security. The quantizer training process is presented in a statistical framework which allows a theoretical analysis of the effects of quantizer training on hashing performance. This is experimentally verified using a number of baseline robust image hashing algorithms over a large database of real world images. This dissertation also proposes a new randomization method for robust image hashing based on Higher Order Spectra (HOS) and Radon projections. The method is non-linear and this is an essential requirement for non-invertibility. The method is also designed to produce features more suited for quantization and encoding. The system can operate without the need for quantizer training, is more easily encoded and displays improved hashing performance when compared to existing robust image hashing algorithms. The dissertation also shows how the HOS method can be adapted to work with biometric features obtained from 2D and 3D face images.

Vision-only autonomous navigation using topometric maps

This paper presents a mapping and navigation system for a mobile robot, which uses vision as its sole sensor modality. The system enables the robot to navigate autonomously, plan paths and avoid obstacles using a vision based topometric map of its environment. The map consists of a globally-consistent pose-graph with a local 3D point cloud attached to each of its nodes. These point clouds are used for direction independent loop closure and to dynamically generate 2D metric maps for locally optimal path planning. Using this locally semi-continuous metric space, the robot performs shortest path planning instead of following the nodes of the graph --- as is done with most other vision-only navigation approaches. The system exploits the local accuracy of visual odometry in creating local metric maps, and uses pose graph SLAM, visual appearance-based place recognition and point clouds registration to create the topometric map. The ability of the framework to sustain vision-only navigation is validated experimentally, and the system is provided as open-source software.

Vision-based place recognition : how low can you go?

In this paper we use the algorithm SeqSLAM to address the question, how little and what quality of visual information is needed to localize along a familiar route? We conduct a comprehensive investigation of place recognition performance on seven datasets while varying image resolution (primarily 1 to 512 pixel images), pixel bit depth, field of view, motion blur, image compression and matching sequence length. Results confirm that place recognition using single images or short image sequences is poor, but improves to match or exceed current benchmarks as the matching sequence length increases. We then present place recognition results from two experiments where low-quality imagery is directly caused by sensor limitations; in one, place recognition is achieved along an unlit mountain road by using noisy, long-exposure blurred images, and in the other, two single pixel light sensors are used to localize in an indoor environment. We also show failure modes caused by pose variance and sequence aliasing, and discuss ways in which they may be overcome. By showing how place recognition along a route is feasible even with severely degraded image sequences, we hope to provoke a re-examination of how we develop and test future localization and mapping systems.