Discriminative Space Research Articles

Discriminative Boosting Algorithm for Diversified Front-End Phonotactic Language Recognition

Currently, phonotactic spoken language recognition (SLR) and acoustic SLR systems are widely used language recognition systems. Parallel phone recognition followed by vector space modeling (PPRVSM) is one typical phonotactic system for spoken language recognition. To achieve better performance, researchers assumed to extract more complementary information of the training data using phone recognizers trained for multiple language-specific phone recognizers, different acoustic models and acoustic features. These methods achieve good performance but usually compute at high computational cost and only using complementary information of the training data. In this paper, we explore a novel approach to discriminative vector space model (VSM) training by using a boosting framework to use the discriminative information of test data effectively, in which an ensemble of VSMs is trained sequentially. The effectiveness of our boosting variation comes from the emphasis on working with the high confidence test data to achieve discriminatively trained models. Our variant of boosting also includes utilizing original training data in VSM training. The discriminative boosting algorithm (DBA) is applied to the National Institute of Standards and Technology (NIST) language recognition evaluation (LRE) 2009 task and show performance improvements. The experimental results demonstrate that the proposed DBA shows 1.8 %, 11.72 % and 15.35 % relative reduction for 30s, 10s and 3s test utterances in equal error rate (EER) than baseline system.

Aplicación de las funciones elípticas de Fourier para la descripción de la forma de los huevos de las aves

Egg shape is difficult to quantify due to the lack of an exact formula to describe its geometry. Here I described a simple algorithm to characterize and compare egg shapes using Fourier functions. These functions can delineate any closed contour and had been previously applied to describe several biological objects. I described, step by step, the process of data acquisition, processing and the use of the SHAPE software to extract function coefficients in a study case. I compared egg shapes in three birds' species representing different reproductive strategies: Cuban Parakeet (Aratinga euops), Royal Tern (Thalasseus maximus) and Cuban Blackbird (Dives atroviolaceus). Using 73 digital pictures of eggs kept in Cuban scientific collections, I calculated Fourier descriptors with 4, 6, 8, 16 and 20 harmonics. Descriptors were reduced by a Principal Component Analysis and the scores of the eigen-values that account for 90% of variance were used in a Lineal Discriminant Function to analyze the possibility to differentiate eggs according to its shapes. Using four harmonics, the first five component accounted for 97% of shape variances; more harmonics diluted the variance increasing to eight the number of components needed to explain most of the variation. Convex polygons in the discriminant space showed a clear separation between species, allowing trustful discrimination (classification errors between 7-15%). Misclassifications were related to specific egg shape variability between species. In the study case, A. euops eggs were perfectly classified, but for the other species, errors ranged from 5 to 29% of misclassifications, in relation to the numbers or harmonics and components used. The proposed algorithm, despite its apparent mathematical complexity, showed many advantages to describe eggs shape allowing a deeper understanding of factors related to this variable.

Snout shape in extant ruminants.

Snout shape is a prominent aspect of herbivore feeding ecology, interacting with both forage selectivity and intake rate. Previous investigations have suggested ruminant feeding styles can be discriminated via snout shape, with grazing and browsing species characterised by ‘blunt’ and ‘pointed’ snouts respectively, often with specification of an ‘intermediate’ sub-grouping to represent ambiguous feeding styles and/or morphologies. Snout shape morphology is analysed here using a geometric morphometric approach to compare the two-dimensional profiles of the premaxilla in ventral aspect for a large sample of modern ruminant species, for which feeding modes are known from secondary criteria. Results suggest that, when browsing and grazing ruminants are classified ecologically based on a range of feeding style indicators, they cannot be discriminated unambiguously on the basis of snout profile shape alone. Profile shapes in our sample form a continuum with substantial overlap between groupings and a diverse range of morphologies. Nevertheless, we obtained an 83.8 percent ratio of correct post hoc feeding style categorisations based on the proximity of projected profile shapes to group centroids in the discriminant space. Accordingly, this procedure for identifying species whose feeding strategy is ‘unknown’ can be used with a reasonable degree of confidence, especially if backed-up by additional information. Based on these results we also refine the definitions of snout shape varieties, taking advantage of the descriptive power that geometric morphometrics offers to characterize the morphological disparities observed. The shape variance exhibited by both browsing and grazing ruminants corresponds strongly to body mass, providing further evidence for an interaction between snout shape, feeding style, and body size evolution. Finally, by exploring the role of phylogenetic similarity in snout shape, we find a slight increase in successful categorisation when repeating the analysis with phylogenetic control on the geometric profiles.

Crop Mapping of Complex Agricultural Landscapes Based on Discriminant Space

Crop mapping is related to planting conditions, remote sensing data source, and performances of various classifiers. Considering the above key factors, this study discusses a crop mapping approach for complex agricultural landscapes based on a conceptual frame that is called discriminant space. The introduced discriminant space-based model follows the pattern and process rules in geosciences and is found by the variables closely related to crop growth. On this basis, a method of classification based on probability calibration algorithm (CPCA) is developed, which consists of several steps. First, discriminant variables were extracted and then the experimental discriminant space was founded. Spectral analysis was carried out in this process to select the dominant impact factors for crop mapping. Based on the founded discriminant space, we calculated the class-conditional probabilities for each category using kernel density estimation and separated the residuals of class-conditional probabilities from the structural component through a trend-surface analysis. Then, calibration of the class-conditional probabilities was implemented using the predicted residuals generated by simple kriging with varying local means. Classification was achieved according to the calibrated probabilities with respect to each category. Accuracy assessment and kappa analysis were used to evaluate the identification results. Statistical test procedure was applied to measure the differences between classification maps and the validation data based on individual categories. Results are discussed for an intensively cropped area in south China. Compared with two other supervised classifiers, the overall accuracy of the proposal method increases by more than 10% and is able to achieve 87.12%. Moreover, there are some substantial increases in the producer's accuracy and the user's accuracy of single-cropping rice and late rice, with a precision increase of more than 15%, effectively improves the identification accuracy of rice in the research region. The proposed method overcomes the limitations due to spectrum characteristics and considers the structural and the random properties of spatial distribution for categories in local regions. Therefore, a similar operation can usually be implemented for crop identification of complex agricultural landscapes.

Fusion Algorithm Based on Improved Linear Local Tangent Space Alignment Algorithm

A uncorrelated adaptive discriminant linear local tangent space alignment (UDALLTSA) is proposed based on improved linear local tangent space alignment algorithm. The algorithm uses an adaptive neighborhood selection to select the appropriate neighborhood, and introduces curvature to amend the model, modifies the constraints of the objective function by use inter-class scatter matrix, and constraints on basis vectors to compute the best projection matrix. By comparing the results of the experiments, it shows that after integrating the discriminant information into the algorithm , uncorrelated constraints and adaptive neighborhood selection can well improve the recognition rate and robustness, thus, possessing good noise immunity, and eliminating redundant information of base vectors, make this fusion algorithm a supervised learning algorithm.

Multi-class ERP-based BCI data analysis using a discriminant space self-organizing map.

Emotional or non-emotional image stimulus is recently applied to event-related potential (ERP) based brain computer interfaces (BCI). Though the classification performance is over 80% in a single trial, a discrimination between those ERPs has not been considered. In this research we tried to clarify the discriminability of four-class ERP-based BCI target data elicited by desk, seal, spider images and letter intensifications. A conventional self organizing map (SOM) and newly proposed discriminant space SOM (ds-SOM) were applied, then the discriminabilites were visualized. We also classify all pairs of those ERPs by stepwise linear discriminant analysis (SWLDA) and verify the visualization of discriminabilities. As a result, the ds-SOM showed understandable visualization of the data with a shorter computational time than the traditional SOM. We also confirmed the clear boundary between the letter cluster and the other clusters. The result was coherent with the classification performances by SWLDA. The method might be helpful not only for developing a new BCI paradigm, but also for the big data analysis.

Learning object-to-class kernels for scene classification.

High-level image representations have drawn increasing attention in visual recognition, e.g., scene classification, since the invention of the object bank. The object bank represents an image as a response map of a large number of pretrained object detectors and has achieved superior performance for visual recognition. In this paper, based on the object bank representation, we propose the object-to-class (O2C) distances to model scene images. In particular, four variants of O2C distances are presented, and with the O2C distances, we can represent the images using the object bank by lower-dimensional but more discriminative spaces, called distance spaces, which are spanned by the O2C distances. Due to the explicit computation of O2C distances based on the object bank, the obtained representations can possess more semantic meanings. To combine the discriminant ability of the O2C distances to all scene classes, we further propose to kernalize the distance representation for the final classification. We have conducted extensive experiments on four benchmark data sets, UIUC-Sports, Scene-15, MIT Indoor, and Caltech-101, which demonstrate that the proposed approaches can significantly improve the original object bank approach and achieve the state-of-the-art performance.

Kernel Reference Discriminant Analysis

Linear Discriminant Analysis (LDA) and its nonlinear version Kernel Discriminant Analysis (KDA) are well-known and widely used techniques for supervised feature extraction and dimensionality reduction. They determine an optimal discriminant space for (non)linear data projection based on certain assumptions, e.g. on using normal distributions (either on the input or in the kernel space) for each class and employing class representation by the corresponding class mean vectors. However, there might be other vectors that can be used for classes representation, in order to increase class discrimination in the resulted feature space. In this paper, we propose an optimization scheme aiming at the optimal class representation, in terms of Fisher ratio maximization, for nonlinear data projection. Compared to the standard approach, the proposed optimization scheme increases class discrimination in the reduced-dimensionality feature space and achieves higher classification rates in publicly available data sets.

Selection of an efficient feature space for EEG-based mental task discrimination

The aim of this paper is to contribute toward exploring an optimal feature space for discriminating mental tasks. Empirical mode decomposition (EMD) algorithm seems useful for designing such a feature space. The adjustment of nonlinear and non-stationary properties of the EEG signals with this algorithm and the successful application of this approach together biomedical signal processing problems encourage us to examine a variety of statistical and spectral measures within the EMD space as the adapted features. In this sense, as a measure of complexity, the Lempel–Ziv algorithm is utilized within the framework of the EMD algorithm. A modified form of the Lempel–Ziv complexity algorithm is then proposed. The features derived from the modified algorithm outperform the other features individually. By combining the modified Lempel–Ziv features with the other adopted features, in average, 97.78% classification accuracy is achieved for different subjects. It is concluded that the EMD–LZ kernel allows for achieving of better performances in classifying mental tasks than the results obtained with other methods.

Point-Manifold Discriminant Analysis for Still-to-Video Face Recognition

In Still-to-Video (S2V) face recognition, only a few high resolution images are registered for each subject, while the probe is video clips of complex variations. As faces present distinct characteristics under different scenarios, recognition in the original space is obviously inefficient. Thus, in this paper, we propose a novel discriminant analysis method to learn separate mappings for different scenario patterns (still, video), and further pursue a common discriminant space based on these mappings. Concretely, by modeling each video as a manifold and each image as point data, we form the scenario-oriented mapping learning as a Point-Manifold Discriminant Analysis (PMDA) framework. The learning objective is formulated by incorporating the intra-class compactness and inter-class separability for good discrimination. Experiments on the COX-S2V dataset demonstrate the effectiveness of the proposed method.

Person Re-Identification by Common-Near-Neighbor Analysis

Re-identifying the same person in different images is a distinct challenge for visual surveillance systems. Building an accurate correspondence between highly variable images requires a suitable dissimilarity measure. To date, most existing measures have used adapted distance based on a learned metric. Unfortunately, real-world human image data, which tends to show large intra-class variations and small inter-class differences, continues to prevent these measures from achieving satisfactory re-identification performance. Recognizing neighboring distribution can provide additional useful information to help tackle the deviation of the to-be-measured samples, we propose a novel dissimilarity measure from the neighborhood-wise relative information perspective, which can deliver the effectiveness of those well-distributed samples to the badly-distributed samples to make intra-class dissimilarities smaller than inter-class dissimilarities, in a learned discriminative space. The effectiveness of this method is demonstrated by explanation and experimentation.

Face relighting using discriminative 2D spherical spaces for face recognition

As part of the face recognition task in a robust security system, we propose a novel approach for the illumination recovery of faces with cast shadows and specularities. Given a single 2D face image, we relight the face object by extracting the nine spherical harmonic bases and the face spherical illumination coefficients by using the face spherical spaces properties. First, an illumination training database is generated by computing the properties of the spherical spaces out of face albedo and normal values estimated from 2D training images. The training database is then discriminately divided into two directions in terms of the illumination quality and light direction of each image. Based on the generated multi-level illumination discriminative training space, we analyze the target face pixels and compare them with the appropriate training subspace using pre-generated tiles. When designing the framework, practical real-time processing speed and small image size were considered. In contrast to other approaches, our technique requires neither 3D face models nor restricted illumination conditions for the training process. Furthermore, the proposed approach uses one single face image to estimate the face albedo and face spherical spaces. In this work, we also provide the results of a series of experiments performed on publicly available databases to show the significant improvements in the face recognition rates.

Learning Discriminative Space–Time Action Parts from Weakly Labelled Videos

Current state-of-the-art action classification methods aggregate space---time features globally, from the entire video clip under consideration. However, the features extracted may in part be due to irrelevant scene context, or movements shared amongst multiple action classes. This motivates learning with local discriminative parts, which can help localise which parts of the video are significant. Exploiting spatio-temporal structure in the video should also improve results, just as deformable part models have proven highly successful in object recognition. However, whereas objects have clear boundaries which means we can easily define a ground truth for initialisation, 3D space---time actions are inherently ambiguous and expensive to annotate in large datasets. Thus, it is desirable to adapt pictorial star models to action datasets without location annotation, and to features invariant to changes in pose such as bag-of-feature and Fisher vectors, rather than low-level HoG. Thus, we propose local deformable spatial bag-of-features in which local discriminative regions are split into a fixed grid of parts that are allowed to deform in both space and time at test-time. In our experimental evaluation we demonstrate that by using local space---time action parts in a weakly supervised setting, we are able to achieve state-of-the-art classification performance, whilst being able to localise actions even in the most challenging video datasets.

Using single-trial EEG to predict and analyze subsequent memory

We show that it is possible to successfully predict subsequent memory performance based on single-trial EEG activity before and during item presentation in the study phase. Two-class classification was conducted to predict subsequently remembered vs. forgotten trials based on subjects' responses in the recognition phase. The overall accuracy across 18 subjects was 59.6% by combining pre- and during-stimulus information. The single-trial classification analysis provides a dimensionality reduction method to project the high-dimensional EEG data onto a discriminative space. These projections revealed novel findings in the pre- and during-stimulus periods related to levels of encoding. It was observed that the pre-stimulus information (specifically oscillatory activity between 25 and 35Hz) −300 to 0ms before stimulus presentation and during-stimulus alpha (7–12Hz) information between 1000 and 1400ms after stimulus onset distinguished between recollection and familiarity while the during-stimulus alpha information and temporal information between 400 and 800ms after stimulus onset mapped these two states to similar values.

Learning a Discriminative Isotropic Space from Labeled and Unlabeled Data

Euclidean distance measure is computationally simple and commonly used in the task of classification. However, it does not capitalize on any discriminant information from training samples, which is crucial to classification performance. In this paper, we learn a Discriminative Isotropic Space from both labeled and unlabeled data to improve classification accuracy as well as generalization ability. Our regularized objective function implicitly minimizes mutual information between input space and transformed space. There is no local optimum problem in our algorithm and its closed form solution can be easily obtained. Experiments on real data sets demonstrate the efficacy of our method.

On the optimal class representation in linear discriminant analysis.

Linear discriminant analysis (LDA) is a widely used technique for supervised feature extraction and dimensionality reduction. LDA determines an optimal discriminant space for linear data projection based on certain assumptions, e.g., on using normal distributions for each class and employing class representation by the mean class vectors. However, there might be other vectors that can represent each class, to increase class discrimination. In this brief, we propose an optimization scheme aiming at the optimal class representation, in terms of Fisher ratio maximization, for LDA-based data projection. Compared with the standard LDA approach, the proposed optimization scheme increases class discrimination in the reduced dimensionality space and achieves higher classification rates in publicly available data sets.

Contribution of a color space selection to a flaw detection vision system

We present a line scan color vision system that detects aspect flaws occurring on the color surfaces of drinking glasses decorated due to an industrial silk-screen process. For this purpose, we have designed a specific image acquisition device based on a color line-scan camera. As the pattern printed on glasses slightly varies between two glasses successively produced, flaw detection by color image analysis is a challenging problem. In order to overcome this problem, the aspect flaws detection is based on an original color image segmentation scheme that iteratively constructs pixel classes. Since the results of segmentation scheme are known to depend on the choice of the color space, the main originality of our approach is to automatically determine the most discriminant color space for each class to be constructed. Experimental results show that the selection of the well-suited color spaces contributes to improve the segmentation accuracy and the aspect flaw detection.

Response of the XENON100 dark matter detector to nuclear recoils

Results from the nuclear recoil calibration of the XENON100 dark matter detector installed underground at the Laboratori Nazionali del Gran Sasso (LNGS), Italy are presented. Data from measurements with an external 241AmBe neutron source are compared with a detailed Monte Carlo simulation which is used to extract the energy dependent charge-yield Qy and relative scintillation efficiency Leff. A very good level of absolute spectral matching is achieved in both observable signal channels - scintillation S1 and ionization S2 - along with agreement in the 2-dimensional particle discrimination space. The results confirm the validity of the derived signal acceptance in earlier reported dark matter searches of the XENON100 experiment.

A fast classification scheme and its application to face recognition

To overcome the high computational complexity in real-time classifier design, we propose a fast classification scheme. A new measure called ‘reconstruction proportion’ is exploited to reflect the discriminant information. A novel space called the ‘reconstruction space’ is constructed according to the reconstruction proportions. A point in the reconstruction space denotes the case of a sample reconstructed using training samples. This is used to search for an optimal mapping from the conventional sample space to the reconstruction space. When the projection from the sample space to the reconstruction space is obtained, a new sample after mapping to the new discriminant space would be classified quickly according to the reconstruction proportions in the reconstruction space. This projection technique results in a diversion of time-consuming calculations from the classification stage to the training stage. Though training time is prolonged, it is advantageous in that classification problems such as identification can be solved in real time. Experimental results on the ORL, Yale, YaleB, and CMU PIE face databases showed that the proposed fast classification scheme greatly outperforms conventional classifiers in classification accuracy and efficiency.

Landmark-based morphometric and meristic variations of Glossogobius giuris in three stocks

Landmark-based morphometric and meristic analysis of pond, haor and estuary populations of Glossogobius giuris (Hamilton,1822) (Perciformes:Gobiidae) were done to find out variations among the three stocks. Samples were collected from three different regions such as pond of Mymensingh, haor of Kishoreganj and estuary of Barisal. Significant differences (*P<0.01) were observed in 13 general morphometric and 23 size adjusted landmark-based morphometric measurements among three stocks. Among the 11 meristic counts the first dorsal fin rays (D1FR), transverse scale above lateral line (TSALL), branchiostegal rays and number of vertebrae were same among fishes of these stocks. In case of pectoral fin rays (PcFR) and transverse scale on lateral line (TSOLL) the haor stock was significantly different from other two stocks. In discriminant space, pond stock was isolated from other two stocks. On the other hand, haor and estuary stocks showed very close relationship. A dendrogram based on the hierarchical cluster analysis using size adjusted general morphometric and landmark measurements formed two main clusterspond stock in one cluster and other two stocks (haor and estuary) remained in another cluster. It indicates that pond stock was separated. The second cluster explained that haor and estuary stocks had very close relationship.DOI: http://dx.doi.org/10.3329/jbau.v10i2.14931 J. Bangladesh Agril. Univ. 10(2): 375-384, 2012