Enhancing consensus in multiple expert decision fusion

ENCORE (Enhanced Consensus in Recognition) is a new classifier structure based on decision fusion of multiple experts (classifiers). When more than one classifier (expert) is available and it is required to combine their decisions, a fundamental aim may be to incorporate a sense of decision consensus. Alternatively, it may be considered important to ensure that appropriate weights are given to more competent classifiers. These two requirements may be mutually contradictory, as the first aims to ensure giving higher emphasis to the best decision delivered by the majority, while the second aims to ensure finding the most appropriate classifier and then giving higher weight to its decision. A new multiple expert classifier (ENCORE) is introduced which implements a decision consensus approach: but the quality of the consensus is evaluated in terms of the past track record of the consenting experts before it is accepted. The ENCORE system has been found to offer greater flexibility of performance in a character recognition task. Detailed analysis using two different databases illustrates the capabilities of this system, although the structure proposed is generic in nature, and may be readily applied to other task domains     

Multiexpert automatic speech recognition using acoustic and myoelectric signals

Classification accuracy of conventional automatic speech recognition (ASR) systems can decrease dramatically under acoustically noisy conditions. To improve classification accuracy and increase system robustness a multiexpert ASR system is implemented. In this system, acoustic speech information is supplemented with information from facial myoelectric signals (MES). A new method of combining experts, known as the plausibility method, is employed to combine an acoustic ASR expert and a MES ASR expert. The plausibility method of combining multiple experts, which is based on the mathematical framework of evidence theory, is compared to the Borda count and score-based methods of combination. Acoustic and facial MES data were collected from 5 subjects, using a 10-word vocabulary across an 18-dB range of acoustic noise. As expected the performance of an acoustic expert decreases with increasing acoustic noise; classification accuracies of the acoustic ASR expert are as low as 11.5%. The effect of noise is significantly reduced with the addition of the MES ASR expert. Classification accuracies remain above 78.8% across the 18-dB range of acoustic noise, when the plausibility method is used to combine the opinions of multiple experts. In addition, the plausibility method produced classification accuracies higher than any individual expert at all noise levels, as well as the highest classification accuracies, except at the 9-dB noise level. Using the Borda count and score-based multiexpert systems, classification accuracies are improved relative to the acoustic ASR expert but are as low as 51.5% and 59.5%, respectively.     

Improved Two‐Phase Framework for Facial Emotion Recognition

Automatic emotion recognition based on facial cues, such as facial action units (AUs), has received huge attention in the last decade due to its wide variety of applications. Current computer‐based automated two‐phase facial emotion recognition procedures first detect AUs from input images and then infer target emotions from the detected AUs. However, more robust AU detection and AU‐to‐emotion mapping methods are required to deal with the error accumulation problem inherent in the multiphase scheme. Motivated by our key observation that a single AU detector does not perform equally well for all AUs, we propose a novel two‐phase facial emotion recognition framework, where the presence of AUs is detected by group decisions of multiple AU detectors and a target emotion is inferred from the combined AU detection decisions. Our emotion recognition framework consists of three major components — multiple AU detection, AU detection fusion, and AU‐to‐emotion mapping. The experimental results on two real‐world face databases demonstrate an improved performance over the previous two‐phase method using a single AU detector in terms of both AU detection accuracy and correct emotion recognition rate.     

多重加权改进型指数双向联想记忆网络及其决策性能

C C Wang等作者利用指数双向联想记忆模型(eBAM),构造了由多个eBAM构成的多重eBAM(Multi-eBAM)信念组合模型,使之可模拟多个专家的投票表决决策,并获得了Multi-eBAM在各eBAM具有同等权威度条件下的决策性能.本文在此基础上,通过对各eBAM引入不同的权值来模拟各专家不同的权威度,推广了Multi-eBAM.进一步借助陈所提出的改进型eBAM(IeBAM),构建了相应的多重加权改进型eBAM(Multi-WIeBAM)信念组合模型,获得了此推理模型在同、异步方式下的决策性能及多专家不同权威度下的多数投票因子,使之更符合实际的多数表决决策.理论分析表明Multi-WIeBAM所获得的多数投票因子优于Multi-WeBAM的多数投票因子,即前者较后者具有更紧致的下界.实验结果也表明了Multi-WIeBAM的性能要优于Multi-WeBAM.     

结合多决策准则稀疏表示的SAR图像目标识别方法

提出组合多决策准则的稀疏表示分类（Sparse Representation-based Classification,SRC）并在合成孔径雷达（Synthetic Aperture Radar,SAR）目标识别中进行应用。传统SRC通常在全局字典上对测试样本进行重构,分别计算不同训练类别对于测试样本的重构误差,最终根据最小重构误差的原则进行分类决策。然而,由于SAR目标识别问题的复杂性,单一决策准则往往对扩展操作条件的适应性不强,导致整体性能下降。为此,文中基于稀疏表示求解的系数矢量,分别采用最小重构误差原则、最大系数能量原则以及局部最小重构误差原则分别进行分类。最小重构误差准则直接采用传统算法。最大系数能量准则分别计算不同训练类别系数能量,按照能量最大的原则进行判决。局部最小重构误差原则在局部字典上对测试样本进行表征和分析,充分体现SAR图像的视角敏感性。对于三个准则获取的决策变量,通过适当转换统一采用概率分布形式进行表达。最终,基于线性加权融合对三个准则的结果进行分析,判决测试样本所属目标类别。基于MSTAR数据集对方法进行测试,分别验证了提出方法在标准操作条件、俯仰角差异、噪声干扰及目标遮挡等情形的性能。实验结果表明:所提方法通过结合多决策准则能够有效提升SAR目标识别性能。     

A framework for fuzzy recognition technology

Presents a scheme for object recognition by classificatory problem solving in the framework of fuzzy sets and possibility theory. The scheme has a particular focus on handling the imperfection problems that are common in application domains where the objects to be recognized (detected and identified) represent undesirable situations, referred to as crises. Crises develop over time, and observations typically increase in number and precision as the crisis develops. Early detection and precise recognition of crises is desired, since it increases the possibility of an effective treatment. The crisis recognition problem is central in several areas of decision support, such as medical diagnosis, financial decision making and early warning systems. The problem is characterized by vague knowledge and observations suffering from several kinds of imperfections, such as missing information, imprecision, uncertainty, unreliability of the source, and mutual (possibly conflicting or reinforcing) observations of the same phenomena. The problem of handling possibly imperfect observations from multiple sources includes the problems of information fusion and multiple-sensor data fusion. The different kinds of imperfection are handled in the framework of fuzzy sets and possibility theory     

Automatic target recognition using a feature-decomposition anddata-decomposition modular neural network

A modular neural network classifier has been applied to the problem of automatic target recognition using forward-looking infrared (FLIR) imagery. The classifier consists of several independently trained neural networks. Each neural network makes a decision based on local features extracted from a specific portion of a target image. The classification decisions of the individual networks are combined to determine the final classification. Experiments show that decomposition of the input features results in performance superior to a fully connected network in terms of both network complexity and probability of classification. Performance of the classifier is further improved by the use of multiresolution features and by the introduction of a higher level neural network on the top of the individual networks, a method known as stacked generalization. In addition to feature decomposition, we implemented a data-decomposition classifier network and demonstrated improved performance. Experimental results are reported on a large set of real FLIR images.     

Classification of anatomical structures in MR brain images using fuzzy parameters

We present an algorithm that automatically segments and classifies the brain structures in a set of magnetic resonance (MR) brain images using expert information contained in a small subset of the image set. The algorithm is intended to do the segmentation and classification tasks mimicking the way a human expert would reason. The algorithm uses a knowledge base taken from a small subset of semiautomatically classified images that is combined with a set of fuzzy indexes that capture the experience and expectation a human expert uses during recognition tasks. The fuzzy indexes are tissue specific and spatial specific, in order to consider the biological variations in the tissues and the acquisition inhomogeneities through the image set. The brain structures are segmented and classified one at a time. For each brain structure the algorithm needs one semiautomatically classified image and makes one pass through the image set. The algorithm uses low-level image processing techniques on a pixel basis for the segmentations, then validates or corrects the segmentations, and makes the final classification decision using higher level criteria measured by the set of fuzzy indexes. We use single-echo MR images because of their high volumetric resolution; but even though we are working with only one image per brain slice, we have multiple sources of information on each pixel: absolute and relative positions in the image, gray level value, statistics of the pixel and its three-dimensional neighborhood and relation to its counterpart pixels in adjacent images. We have validated our algorithm for ease of use and precision both with clinical experts and with measurable error indexes over a Brainweb simulated MR set.     

Adaptive Algorithm of Tracking the Best Experts Trajectory

The problem of decision theoretic online learning is discussed. There is the set of methods, experts, and algorithms capable of making solutions (or predictions) and suffering losses due to the inaccuracy of their solutions. An adaptive algorithm whereby expert solutions are aggregated and sustained losses not exceeding (to a certain quantity called a regret) those of the best combination of experts distributed over the prediction interval is proposed. The algorithm is constructed using the Fixed-Share method combined with the Ada-Hedge algorithm used to exponentially weight expert solutions. The regret of the proposed algorithm is estimated. In the context of the given approach, there are no any stochastic assumptions about an initial data source and the boundedness of losses. The results of numerical experiments concerning the mixing of expert solutions with the help of the proposed algorithm are presented. The strategies of games on financial markets, which were suggested in our previous papers, play the role of expert strategies.     

Complex character decomposition using deformable model

Despite the fact that Chinese characters are composed of radicals and Chinese people usually formulate their knowledge of Chinese characters as a combination of radicals, very few studies have focused on a character decomposition approach to recognition, i.e. recognizing a character by first extracting and recognizing its radicals. In this paper, such an approach is adopted, and the problem of how to extract radical sub-images from character images is addressed by proposing an algorithm based on a deformable model (DM). The application of a DM to complex character decomposition (and recognition) is a novel one, and concepts like goodness of character decomposition have been exploited to formulate appropriate energy terms and to devise cost-effective minimization schemes for the problem. The advantage of the character decomposition approach is demonstrated by feeding the extracted radical images to an existing structure-based Chinese character recognizer, the outputs of which are then combined to classify the input. Simulation results show that the performance of the existing system can be improved significantly when character decomposition is used     

基于直觉语言评价的电力变压器状态维修决策方法

研究电力变压器状态维修的决策方法对提高其运行安全和可靠性具有重要意义。针对当前变压器状态维修决策方法对定性评价信息处理的不足,文中结合直觉模糊集和语言评价法提出一种基于直觉语言评价的决策方法。为得到更为完备的决策信息,该方法首先引入直觉语言数形式以充分表达专家评语的肯定度、犹豫度;在此基础上,基于直觉语言熵定义并综合考虑多方面因素建立起优化模型来确定指标权值,同时利用直觉二元语言算子对专家群体决策信息进行集成运算,以进行方案优选。实例分析表明,该方法能涵盖更多的决策信息,并减小定性到定量转化过程中的信息量损失,使得决策结果更加客观、合理和全面。     

Acquisition for DS/CDMA systems with multiple antennas in frequency-selective fading channels

A generalized code acquisition scheme for direct-sequence code-division multiple-access systems with multiple antennas is proposed over frequency-selective fading channels. The proposed scheme is developed on the framework of a generalized configuration of multiple antennas and correlators. The nonconsecutive search method is generalized and extended to multiple antenna systems to exploit multipath signals in improving acquisition performance over frequency-selective fading channels. The proposed scheme also adopts a hybrid decision strategy to make effective decisions on acquisition. The mean acquisition time performance of the proposed acquisition scheme is analyzed and evaluated in frequency-selective Rayleigh-fading channels with general multipath delay profiles and spatial-fading correlations. The effects of nonconsecutive search on mean acquisition time are investigated for various channel environments, and the optimal choice of decision strategy is discussed. Furthermore, effects of various configurations of multiple antennas and correlators, decision thresholds, and correlation interval on the performance are also investigated.     

Performance-based classifier combination in atlas-based image segmentation using expectation-maximization parameter estimation

It is well known in the pattern recognition community that the accuracy of classifications obtained by combining decisions made by independent classifiers can be substantially higher than the accuracy of the individual classifiers. We have previously shown this to be true for atlas-based segmentation of biomedical images. The conventional method for combining individual classifiers weights each classifier equally (vote or sum rule fusion). In this paper, we propose two methods that estimate the performances of the individual classifiers and combine the individual classifiers by weighting them according to their estimated performance. The two methods are multiclass extensions of an expectation-maximization (EM) algorithm for ground truth estimation of binary classification based on decisions of multiple experts (Warfield et al., 2004). The first method performs parameter estimation independently for each class with a subsequent integration step. The second method considers all classes simultaneously. We demonstrate the efficacy of these performance-based fusion methods by applying them to atlas-based segmentations of three-dimensional confocal microscopy images of bee brains. In atlas-based image segmentation, multiple classifiers arise naturally by applying different registration methods to the same atlas, or the same registration method to different atlases, or both. We perform a validation study designed to quantify the success of classifier combination methods in atlas-based segmentation. By applying random deformations, a given ground truth atlas is transformed into multiple segmentations that could result from imperfect registrations of an image to multiple atlas images. In a second evaluation study, multiple actual atlas-based segmentations are combined and their accuracies computed by comparing them to a manual segmentation. We demonstrate in both evaluation studies that segmentations produced by combining multiple individual registration-based segmentations are more accurate for the two classifier fusion methods we propose, which weight the individual classifiers according to their EM-based performance estimates, than for simple sum rule fusion, which weights each classifier equally.     

Neural network models of sensory integration for improved vowelrecognition

It is demonstrated that multiple sources of speech information can be integrated at a subsymbolic level to improve vowel recognition. Feedforward and recurrent neural networks are trained to estimate the acoustic characteristics of a vocal tract from images of the speaker's mouth. These estimates are then combined with the noise-degraded acoustic information, effectively increasing the signal-to-noise ratio and improving the recognition of these noise-degraded signals. Alternative symbolic strategies such as direct categorization of the visual signals into vowels are also presented. The performances of these neural networks compare favorably with human performance and with other pattern-matching and estimation techniques     

一种基于多分类器融合的双色红外成像目标识别方法

针对双色红外成像系统中的自动目标识别问题,提出了一种基于多分类器融合的红外目标识别方法.该方法首先提取目标的形状特征和面貌特征,并设计多个基于不同特征的分类器对目标进行分类;然后对各个分类器的目标分类结果进行决策级融合处理,并采用所提出的决策规则对多分类器融合分类结果进行处理得到最终的目标识别结果.该方法充分利用了目标在多传感器图像中的多种分类特征信息,提高了系统的目标识别效率和精确性.实验结果证实了该方法的有效性.     

Multichannel fusion models for the parametric classification of differential brain activity

This paper introduces parametric multichannel fusion models to exploit the different but complementary brain activity information recorded from multiple channels in order to accurately classify differential brain activity into their respective categories. A parametric weighted decision fusion model and two parametric weighted data fusion models are introduced for the classification of averaged multichannel evoked potentials (EPs). The decision fusion model combines the independent decisions of each channel classifier into a decision fusion vector and a parametric classifier is designed to determine the EP class from the discrete decision fusion vector. The data fusion models include the weighted EP-sum model in which the fusion vector is a linear combination of the multichannel EPs and the EP-concatenation model in which the fusion vector is a vector-concatenation of the multichannel EPs. The discrete Karhunen-Loeve transform (DKLT) is used to select features for each channel classifier and from each data fusion vector. The difficulty in estimating the probability density function (PDF) parameters from a small number of averaged EPs is identified and the class conditional PDFs of the feature vectors of averaged EPs are, therefore, derived in terms of the PDFs of the single-trial EPs. Multivariate parametric classifiers are developed for each fusion strategy and the performances of the different strategies are compared by classifying 14-channel EPs collected from five subjects involved in making explicit match/mismatch comparisons between sequentially presented stimuli. It is shown that the performance improves by incorporating weights in the fusion rules and that the best performance is obtained using multichannel EP concatenation. It is also noted that the fusion strategies introduced are also applicable to other problems involving the classification of multicategory multivariate signals generated from multiple sources.     

用于异质信息的信任区间交互式多属性识别方法

为了解决混合类型数据与专家知识等异质信息的融合决策问题,该文提出了基于信任区间的交互式多属性识别(BI-TODIM)方法。完善了混合类型数据的距离测度,根据信任区间的构建定理和灰关联方法构建了未知目标混合类型数据的信任区间,阐明了信任区间与直觉模糊数之间的等价关系,创建了混合类型数据和专家知识的识别决策模型,实现了特征层信息和决策层信息的统一表达;分析了基于信度函数的逼近理想解(BF-TOPSIS)方法的反转现象及算法的复杂度,定义了区间数的序关系,提出了BI-TODIM识别决策方法,及基于直觉模糊熵的未知权重计算方法。结合算例和目标识别案例,验证了该文方法在解决排序反转和异质信息融合方面的有效性,突出了该方法时间复杂度低、稳定性好、识别准确度高的优点。     

Frame-synchronous noise compensation for hands-free speechrecognition in car environments

It has become increasingly important to develop hands-free speech recognition techniques for the human-computer interface in car environments. However, severe car noise degrades the speech recognition performance substantially. To compensate the performance loss, it is necessary to adapt the original speech hidden Markov models (HMMs) to meet changing car environments. A novel frame-synchronous adaptation mechanism for in-car speech recognition is presented. This mechanism is intended to perform unsupervised model adaptation efficiently on a frame-by-frame basis instead of a conventional adaptation algorithm relying on batch adaptation data and supervision information. The proposed adaptation scheme is performed during frame likelihood calculation where an optimal equalisation factor is first computed to equalise the model mean vector and the input frame vector. This equalisation factor then serves as a reference index to retrieve an additional bias vector for model mean adaptation. As a result, a rapid and flexible algorithm is exploited to establish a new robust likelihood measure. In experiments on hands-free in-car speech recognition with the microphone far from the talker, this framework is found to be effective in terms of recognition rate and computational cost under various driving speeds     

用于异质信息的信任区间交互式多属性识别方法

为了解决混合类型数据与专家知识等异质信息的融合决策问题,该文提出了基于信任区间的交互式多属性识别(BI-TODIM)方法。完善了混合类型数据的距离测度,根据信任区间的构建定理和灰关联方法构建了未知目标混合类型数据的信任区间,阐明了信任区间与直觉模糊数之间的等价关系,创建了混合类型数据和专家知识的识别决策模型,实现了特征层信息和决策层信息的统一表达;分析了基于信度函数的逼近理想解(BF-TOPSIS)方法的反转现象及算法的复杂度,定义了区间数的序关系,提出了BI-TODIM识别决策方法,及基于直觉模糊熵的未知权重计算方法。结合算例和目标识别案例,验证了该文方法在解决排序反转和异质信息融合方面的有效性,突出了该方法时间复杂度低、稳定性好、识别准确度高的优点。