In this paper, we present an automatic vision-based traffic sign recognition system, which can detect and classify traffic signs at long distance under different lighting conditions. To realize this purpose, the traffic sign recognition is developed in an originally proposed dual-focal active camera system. In this system, a telephoto camera is equipped as an assistant of a wide angle camera. The telephoto camera can capture a high accuracy image for an object of interest in the view field of the wide angle camera. The image from the telephoto camera provides enough information for recognition when the accuracy of traffic sign is low from the wide angle camera. In the proposed system, the traffic sign detection and classification are processed separately for different images from the wide angle camera and telephoto camera. Besides, in order to detect traffic sign from complex background in different lighting conditions, we propose a type of color transformation which is invariant to light changing. This color transformation is conducted to highlight the pattern of traffic signs by reducing the complexity of background. Based on the color transformation, a multi-resolution detector with cascade mode is trained and used to locate traffic signs at low resolution in the image from the wide angle camera. After detection, the system actively captures a high accuracy image of each detected traffic sign by controlling the direction and exposure time of the telephoto camera based on the information from the wide angle camera. Moreover, in classification, a hierarchical classifier is constructed and used to recognize the detected traffic signs in the high accuracy image from the telephoto camera. Finally, based on the proposed system, a set of experiments in the domain of traffic sign recognition is presented. The experimental results demonstrate that the proposed system can effectively recognize traffic signs at low resolution in different lighting conditions.

Object tracking is a major technique in image processing and computer vision. Tracking speed will directly determine the quality of applications. This paper presents a parallel implementation for a recently proposed scale- and rotation-invariant on-line object tracking system. The algorithm is based on NVIDIA's Graphics Processing Units (GPU) using Compute Unified Device Architecture (CUDA), following the model of single instruction multiple threads. Specifically, we analyze the original algorithm and propose the GPU-based parallel design. Emphasis is placed on exploiting the data parallelism and memory usage. In addition, we apply optimization technique to maximize the utilization of NVIDIA's GPU and reduce the data transfer time. Experimental results show that our GPGPU-based method running on a GTX480 graphics card could achieve up to 12X speed-up compared with the efficiency equivalence on an Intel E8400 3.0 GHz CPU, including I/O time.

This paper investigates a new method for fusion of scores generated by multiple classification sub-systems that help to further reduce the classification error rate in Spoken Language Recognition (SLR). In recent studies, a variety of effective classification algorithms have been developed for SLR. Hence, it has been a common practice in the National Institute of Standards and Technology (NIST) Language Recognition Evaluations (LREs) to fuse the results from several classification sub-systems to boost the performance of the SLR systems. In this work, we introduce a discriminative performance measure to optimize the performance of the fusion of 7 language classifiers developed as IIR's submission to the 2009 NIST LRE. We present an Error Corrective Fusion (ECF) method in which we iteratively learn the fusion weights to minimize error rate of the fusion system. Experiments conducted on the 2009 NIST LRE corpus demonstrate a significant improvement compared to individual sub-systems. Comparison study is also conducted to show the effectiveness of the ECF method.

Boosting over weak classifiers is widely used in pedestrian detection. As the number of weak classifiers is large, researchers always use a sampling method over weak classifiers before training. The sampling makes the boosting process harder to reach the fixed target. In this paper, we propose a partial derivative guidance for weak classifier mining method which can be used in conjunction with a boosting algorithm. Using weak classifier mining method makes the sampling less degraded in the performance. It has the same effect as testing more weak classifiers while using acceptable time. Experiments demonstrate that our algorithm can process quicker than [1] algorithm in both training and testing, without any performance decrease. The proposed algorithms is easily extending to any other boosting algorithms using a window-scanning style and HOG-like features.

The least-squares probabilistic classifier (LSPC) is a computationally-efficient alternative to kernel logistic regression. However, to assure its learned probabilities to be non-negative, LSPC involves a post-processing step of rounding up negative parameters to zero, which can unexpectedly influence classification performance. In order to mitigate this problem, we propose a simple alternative scheme that directly rounds up the classifier's negative outputs, not negative parameters. Through extensive experiments including real-world image classification and audio tagging tasks, we demonstrate that the proposed modification significantly improves classification accuracy, while the computational advantage of the original LSPC remains unchanged.

This paper proposes a self-taught classifier of gateways for hybrid SLAM. Gateways are detected and recognized by the self-taught classifier, which is a SVM classifier and self-taught in that its training samples are produced and labeled without user's intervention. Since the detection of gateways at the topological boundaries of an acquired metric map reduces computational complexity in partitioning the metric map into sub-maps as compared with previous hybrid SLAM approaches using spectral clustering methods, from O(2n) to O(n), where n is the number of sub-maps. This makes possible real time hybrid SLAM even for large-scale metric maps. We have confirmed that the self-taught classifier provides satisfactory consistency and computationally efficiency in hybrid SLAM through different experiments.

This paper presents a novel algorithm for Nearest Neighbor (NN) classifier. NN classification is a well-known method of pattern classification having the following properties: * it performs maximum-margin classification and achieves less than twice the ideal Bayesian error, * it does not require knowledge of pattern distributions, kernel functions or base classifiers, and * it can naturally be applied to multiclass classification problems. Among the drawbacks are A) inefficient memory use and B) ineffective pattern classification speed. This paper deals with the problems A and B. In most cases, NN search algorithms, such as k-d tree, are employed as a pattern search engine of the NN classifier. However, NN classification does not always require the NN search. Based on this idea, we propose a novel algorithm named k-d decision tree (KDDT). Since KDDT uses Voronoi-condensed prototypes, it consumes less memory than naive NN classifiers. We have confirmed that KDDT is much faster than NN search-based classifier through a comparative experiment (from 9 to 369 times faster than NN search based classifier). Furthermore, in order to extend applicability of the KDDT algorithm to high-dimensional NN classification, we modified it by incorporating Gabriel editing or RNG editing instead of Voronoi condensing. Through experiments using simulated and real data, we have confirmed the modified KDDT algorithms are superior to the original one.

This paper proposes the sleep stage estimation method that can provide an accurate estimation for each person without connecting any devices to human's body. In particular, our method learns the appropriate multiple band-pass filters to extract the specific wave pattern of heartbeat, which is required to estimate the sleep stage. For an accurate estimation, this paper employs Learning Classifier System (LCS) as the data-mining techniques and extends it to estimate the sleep stage. Extensive experiments on five subjects in mixed health confirm the following implications: (1) the proposed method can provide more accurate sleep stage estimation than the conventional method, and (2) the sleep stage estimation calculated by the proposed method is robust regardless of the physical condition of the subject.

To prevent blindness from diabetic retinopathy, periodic screening and early diagnosis are neccessary. Due to lack of expert ophthalmologists in rural area, automated early exudate (one of visible sign of diabetic retinopathy) detection could help to reduce the number of blindness in diabetic patients. Traditional automatic exudate detection methods are based on specific parameter configuration, while the machine learning approaches which seems more flexible may be computationally high cost. A comparative analysis of traditional and machine learning of exudates detection, namely, mathematical morphology, fuzzy c-means clustering, naive Bayesian classifier, Support Vector Machine and Nearest Neighbor classifier are presented. Detected exudates are validated with expert ophthalmologists' hand-drawn ground-truths. The sensitivity, specificity, precision, accuracy and time complexity of each method are also compared.

A system for detecting fundus lesions caused by diabetic retinopathy from fundus images is being developed. The system can screen the images in advance in order to reduce the inspection workload on doctors. One of the difficulties that must be addressed in completing this system is how to remove false positives (which tend to arise near blood vessels) without decreasing the detection rate of lesions in other areas. To overcome this difficulty, we developed classifier selection according to the position of a candidate lesion, and we introduced new features that can distinguish true lesions from false positives. A system incorporating classifier selection and these new features was tested in experiments using 55 fundus images with some lesions and 223 images without lesions. The results of the experiments confirm the effectiveness of the proposed system, namely, degrees of sensitivity and specificity of 98% and 81%, respectively.

This paper proposes a novel algorithm for image feature extraction-the dual two-dimensional fuzzy class preserving projections ((2D)2FCPP). The main advantages of (2D)2FCPP over two-dimensional locality preserving projections (2DLPP) are: (1) utilizing the fuzzy assignation mechanisms to construct the weight matrix, which can improve the classification results; (2) incorporating 2DLPP and alternative 2DLPP to get a more efficient dimensionality reduction method-(2D)2LPP.

This paper describes a vision-based and real-time system for detecting road signs from within a moving vehicle. The system architecture which is proposed in this paper consists of two parts, the learning and the detection part of road sign images. The proposed system has the standard architecture with adaboost algorithm. Adaboost is a popular algorithm which used to detect an object in real time. To improve the detection rate of adaboost algorithm, this paper proposes a new combination method of classifiers in every stage. In the case of detecting road signs in real environment, it can be ambiguous to decide to which class input images belong. To overcome this problem, we propose a method that applies fuzzy measure and fuzzy integral which use the importance and the evaluated values of classifiers within one stage. It is called fuzzy-boosting in this paper. Also, to improve the speed of a road sign detection algorithm using adaboost at the detection step, we propose a method which chooses several candidates by using MC generator. In this paper, as the sub-windows of chosen candidates pass classifiers which are made from fuzzy-boosting, we decide whether a road sign is detected or not. Using experiment result, we analyze and compare the detection speed and the classification error rate of the proposed algorithm applied to various environment and condition.

Many studies of on-line comprehension of semantic violations have shown that the human sentence processor rapidly constructs a higher-order semantic interpretation of the sentence. What remains unclear, however, is the amount of time required to detect semantic anomalies while concatenating two words to form a phrase with very rapid stimuli presentation. We aimed to examine the time course of semantic integration in concatenating two words in phrase structure building, using magnetoencephalography (MEG). In the MEG experiment, subjects decided whether two words (a classifier and its corresponding noun), presented each for 66 ms, form a semantically correct noun phrase. Half of the stimuli were matched pairs of classifiers and nouns. The other half were mismatched pairs of classifiers and nouns. In the analysis of MEG data, there were three primary peaks found at approximately 25 ms (M1), 170 ms (M2) and 250 ms (M3) after the presentation of the target words. As a result, only the M3 latencies were significantly affected by the stimulus conditions. Thus, the present results indicate that the semantic integration in concatenating two words starts from approximately 250 ms.

A family of linear subspace classifiers called local subspace classifier (LSC) outperforms the k-nearest neighbor rule (kNN) and conventional subspace classifiers in handwritten digit classification. However, LSC suffers very high sensitivity to image transformations because it uses projection and the Euclidean distances for classification. In this paper, I present a combination of a local subspace classifier (LSC) and a tangent distance (TD) for improving accuracy of handwritten digit recognition. In this classification rule, we can deal with transform-invariance easily because we are able to use tangent vectors for approximation of transformations. However, we cannot use tangent vectors in other type of images such as color images. Hence, kernel LSC (KLSC) is proposed for incorporating transform-invariance into LSC via kernel mapping. The performance of the proposed methods is verified with the experiments on handwritten digit and color image classification.

In this paper, we present a method to automatically acquire a large-scale vocabulary of evaluative expressions from a large corpus of blogs. For the purpose, this paper presents a semi-supervised method for classifying evaluative expressions, that is, tuples of subjects, their attributes, and evaluative words, that indicate either favorable or unfavorable opinions towards a specific subject. Due to its characteristics, our semi-supervised method can classify evaluative expressions in a corpus by their polarities, starting from a very small set of seed training examples and using contextual information in the sentences the expressions belong to. Our experimental results with real Weblog data as our corpus show that this bootstrapping approach can improve the accuracy of methods for classifying favorable and unfavorable opinions. We also show that a reasonable amount of evaluative expressions can be really acquired.

We propose, in this article, the Hierarchical Behavior-Knowledge Space as an extension of Behavior-Knowledge Space. Hierarchical BKS utilizes ranked level individual classifiers, and automatically expands its behavioral knowledge in order to satisfy given reliability requirement. From the statistical view point, its decisions are as optimal as those of original BKS, and the reliability threshold is a lower bound of estimated reliability. Several comparisons with original BKS and unanimous voting are shown with some experiments.

In this paper, we propose a learning classifier based on maximum entropy (ME) for resolving zero-anaphora in Chinese text. Besides regular grammatical, lexical, positional and semantic features motivated by previous research on anaphora resolution, we develop two innovative Web-based features for extracting additional semantic information from the Web. The values of the two features can be obtained easily by querying the Web using some patterns. Our study shows that our machine learning approach is able to achieve an accuracy comparable to that of state-of-the-art systems. The Web as a knowledge source can be incorporated effectively into the ME learning framework and significantly improves the performance of our approach.

In electronic warfare support systems, the analysis of PRI (Pulse Repetition Interval) modulation characteristics for a radar pulse signal has attracted much interest because of the problem of the identification ambiguity in dense electronic warfare signal environments. A new method of recognizing the PRI modulation type of a radar pulse signal is proposed for electronic warfare support. The proposed method recognizes the PRI modulation types using classifiers based on the property of the autocorrelation of the PRI sequences for each PRI modulation type. In addition, the proposed method estimates the PRI modulation period for the PRI modulation type with the periodicity. Simulation results are presented to show the performance of the proposed method.

A multi-stage approach -- which is fast, robust and easy to train -- for a face-detection system is proposed. Motivated by the work of Viola and Jones [1], this approach uses a cascade of classifiers to yield a coarse-to-fine strategy to reduce significantly detection time while maintaining a high detection rate. However, it is distinguished from previous work by two features. First, a new stage has been added to detect face candidate regions more quickly by using a larger window size and larger moving step size. Second, support vector machine (SVM) classifiers are used instead of AdaBoost classifiers in the last stage, and Haar wavelet features selected by the previous stage are reused for the SVM classifiers robustly and efficiently. By combining AdaBoost and SVM classifiers, the final system can achieve both fast and robust detection because most non-face patterns are rejected quickly in earlier layers, while only a small number of promising face patterns are classified robustly in later layers. The proposed multi-stage-based system has been shown to run faster than the original AdaBoost-based system while maintaining comparable accuracy.

Active Shape Model (ASM) has been shown to be a powerful tool to aid the interpretation of images, especially in face alignment. ASM local appearance model parameter estimation is based on the assumption that residuals between model fit and data have a Gaussian distribution. Moreover, to generate an allowable face shape, ASM truncates coefficients of shape principal components into the bounds determined by eigenvalues. In this paper, an algorithm of modeling local appearances, called AdaBoosted ASM, and a shape parameter optimization method are proposed. In the algorithm of modeling the local appearances, we describe our novel modeling method by using AdaBoosted histogram classifiers, in which the assumption of the Gaussian distribution is not necessary. In the shape parameter optimization, we describe that there is an inadequacy on controlling shape parameters in ASM, and our novel method on how to solve it. Experimental results demonstrate that the AdaBoosted histogram classifiers improve robustness of landmark displacement greatly, and the shape parameter optimization solves the inadequacy problem of ASM on shape constraint effectively.