Xiaoping Lai

Extreme learning machine and adaptive sparse representation for image classification

Recent research has shown the speed advantage of extreme learning machine (ELM) and the accuracy advantage of sparse representation classification (SRC) in the area of image classification. Those two methods, however, have their respective drawbacks, e.g., in general, ELM is known to be less robust to noise while SRC is known to be time-consuming. Consequently, ELM and SRC complement each other in computational complexity and classification accuracy. In order to unify such mutual complementarity and thus further enhance the classification performance, we propose an efficient hybrid classifier to exploit the advantages of ELM and SRC in this paper. More precisely, the proposed classifier consists of two stages: first, an ELM network is trained by supervised learning. Second, a discriminative criterion about the reliability of the obtained ELM output is adopted to decide whether the query image can be correctly classified or not. If the output is reliable, the classification will be performed by ELM; otherwise the query image will be fed to SRC. Meanwhile, in the stage of SRC, a sub-dictionary that is adaptive to the query image instead of the entire dictionary is extracted via the ELM output. The computational burden of SRC thus can be reduced. Extensive experiments on handwritten digit classification, landmark recognition and face recognition demonstrate that the proposed hybrid classifier outperforms ELM and SRC in classification accuracy with outstanding computational efficiency.

Optimal Design of Nonlinear-Phase FIR Filters With Prescribed Phase Error

Constrained least-squares design and constrained Chebyshev design of one- and two-dimensional nonlinear-phase FIR filters with prescribed phase error are considered in this paper by a unified semi-infinite positive-definite quadratic programming approach. In order to obtain unique optimal solutions, we propose to impose constraints on the complex approximation error and the phase error. By introducing a sigmoid phase-error constraint bound function, the group-delay error can be greatly reduced. A Goldfarb-Idnani based algorithm is presented to solve the semi-infinite positive-definite quadratic program resulting from the constrained least-squares design problem, and then applied after some modifications to the constrained Chebyshev design problem, which is proved in this paper to be equivalent also to a semi-infinite positive-definite quadratic program. Through design examples, the proposed method is compared with several existing methods. Simulation results demonstrate the effectiveness and efficiency of the proposed method.

Landmark recognition with sparse representation classification and extreme learning machine

Abstract Along with the rapid development of intelligent mobile terminals, applications on landmark recognition attract increasingly attentions by world wide researchers in the past several years. Although promising achievements have been presented, designing a robust recognition system with an accurate recognition rate and fast response speed is still challenging. To address these issues, we propose a novel landmark recognition algorithm in this paper using the spatial pyramid kernel based bag-of-words (SPK-BoW) histogram approach with the feedforward artificial neural networks (FNN) and the sparse representation classifier (SRC). In the proposed algorithm, the SPK-BoW approach is first employed to extract features and construct an overcomplete dictionary for landmark image representation. Then, the FNN trained with the extreme learning machine (ELM) algorithm combined with the SRC is implemented for landmark image recognition. We conduct experiments using the Nanyang Technological University (NTU) campus landmark database to show that the proposed method achieves a high recognition rate than ELM and a lower response time than the sparse representation technique.

Ensemble extreme learning machine and sparse representation classification

Abstract Extreme learning machine (ELM) combining with sparse representation classification (ELM-SRC) has been developed for image classification recently. However, employing a single ELM network with random hidden parameters may lead to unstable generalization and data partition performance in ELM-SRC. To alleviate this deficiency, we propose an enhanced ensemble based ELM and SRC algorithm (En-SRC) in this paper. Rather than using the output of a single ELM to decide the threshold for data partition, En-SRC incorporates multiple ensembles to enhance the reliability of the classifier. Different from ELM-SRC, a theoretical analysis on the data partition threshold selection of En-SRC is given. Extension to the ensemble based regularized ELM with SRC (EnR-SRC) is also presented in the paper. Experiments on a number of benchmark classification databases show that the proposed methods win a better classification performance with a lower computational complexity than the ELM-SRC approach.

Minimax Design of IIR Digital Filters Using a Sequential Constrained Least-Squares Method

The minimax design of infinite impulse response (IIR) digital filters is a nonconvex optimization problem, and thus has many local minima. It is shown in this correspondence that a sequential constrained least-squares (SCLS) method has a higher possibility of obtaining better solutions than a direct minimization method when applied to the nonconvex minimax design of IIR filters. We combine the SCLS method with a Steiglitz-McBride (SM) strategy, resulting in a practical design procedure. The positive realness stability condition proposed by Dumitrescu and Niemisto is reformulated as linear inequality constraints and then incorporated in the design procedure. Simulation examples and comparisons with several existing methods demonstrate the effectiveness of the procedure and good performances of the designed filters.

A Sequential Minimization Procedure for Minimax Design of IIR Filters Based on Second-Order Factor Updates

In order to make good use of the linearity, necessity, and sufficiency of the stability-triangle-based stability conditions, a sequential minimization procedure is proposed in this brief to convert the minimax design of an infinite impulse response (IIR) filter to a sequence of minimax subproblems. Aside from the numerator, only one second-order factor of the denominator is optimized in each of the minimax subproblems. The sequential minimization procedure is then combined with a Levy-Sanathanan-Koerner-based direct minimization method for the subproblems. Many IIR filters have been designed using the proposed procedure. Three examples are provided for a comparison of the procedure with some recent design methods. Design results show that the proposed procedure has obtained much better minimax filters.

Minimax Phase Error Design of IIR Digital Filters With Prescribed Magnitude and Phase Responses

Infinite impulse response (IIR) digital filters with prescribed magnitude and phase responses have been used in many applications. To approximate the prescribed magnitude and phase responses, we propose a new approach to the design of general IIR filters by minimizing the maximum phase error subject to a prescribed or simultaneously minimized maximum magnitude error, where the phase error and magnitude error are controlled by two elliptic constraints respectively with major and minor axes along the desired frequency response. The sequential constrained least-squares method and Levy-Sanathanan-Koerner strategy are used to convert the nonconvex constraints into convex ones, resulting in a series of convex optimization subproblems. Design examples and comparisons with recent methods demonstrate the flexibility and effectiveness of the proposed methods.

Acoustic vector sensor: reviews and future perspectives

Acoustic vector sensor (AVS) has been recently researched and developed for acoustic wave capturing and signal processing. Conventional array generally employs spatially displayed sensors for signal enhancement, source localisation, target tracking, etc. However, the large size usually limits its implementations on some portable devices. AVS which generally includes one omni-directional sensor and three orthogonally co-located directional sensors has been recently introduced. An AVS is able to provide the four-dimensional information of sound field in space: the acoustic pressure and its three-dimensional particle velocities. A compact assembled AVS could be as small as a match head and the weight can be <;50 g. Benefits from these properties, AVS tends to be more attractive for exploitation and commercialisation than conventional sensor array. To have a well understanding of the research progress on AVS, an overview on its recent developments is first given in this study. Then, discussions of challenges on AVS and extensions on its possible future prospects are presented.

Voting based weighted online sequential extreme learning machine for imbalance multi-class classification

In this paper, a voting based weighted online sequential extreme learning machine (VWOS-ELM) is proposed for class imbalance learning (CIL). VWOS-ELM is the first sequential classifier that can tackle the class imbalance problem in multi-class data streams. Utilizing WOS-ELM and the recently proposed voting based online sequential extreme learning machine (VOS-ELM) method, VWOS-ELM adapts better to newly received data than the original WOS-ELM method. Experimental results show that VWOS-ELM outperforms both the WOS-ELM and the recent meta-cognitive extreme learning machine methods. It also achieves similar performance to that of ensemble of subset OS-ELM (ESOS-ELM) but using fewer independent classifiers.

Optimal Design of Constrained FIR Filters Without Phase Response Specifications

Finite impulse response (FIR) filters without phase response specifications have found many applications in signal processing and communications. It is shown in this paper that the constrained Lp magnitude error design of such an FIR filter is equivalent to a constrained Lp approximation of a given response with a phase equal to that of the optimal filter of the original problem. An iterative method is then proposed to compute that phase by converting the nonconvex constrained Lp magnitude error problem into a series of convex constrained Lp elliptic frequency response error subproblems. It is also shown that the solutions of these convex subproblems converge to a Karush-Kuhn-Tucker point of the original problem. The convergence and its model parameter and initial condition dependence are shown through the designs of FIR filters without time-domain constraints. The iterative method is then applied to the minimax design of evidence filters, Nyquist filters, and step response constrained filters, and to the Lp design of pulse shaping filters for ultra-wideband systems. Design examples demonstrate that the proposed method obtains better filters in terms of magnitude error, group delay, and spectrum utilization efficiency than existing methods.