Search

In this paper, a new nonlinear feature extraction method based on the WTMM (wavelet transform modulus-maxima method) is proposed, which can greatly facilitate the extraction of the multifractal spectrum feature (MSF) from speech signals. The MSF combined with traditional linear features can obviously improve the performance of speaker recognition system. Experiment results show that 6-dimensional MSF combined with LPC make recognition accuracy increase 6.4 percentage points, and 6-dimensional MSF combined with MFCC, LPC make recognition accuracy increase 1.6 percentage points and reach 98.8% in short speech (2 seconds) speaker recognition.

A novel rate control scheme for H.264/SVC base layer is presented in this paper. At first, a new mean-absolute-difference (MAD) estimation method is introduced to replace the traditional MAD linear prediction model. This new method could reduce the computation complexity sharply and decrease the MAD estimation error by up to 35.6%. Then, the quadratic rate quantization (R-Q) model and the linear R-Q model from [1] are introduced to describe the relation between the sum of texture bits and non-texture bits and the quantization parameters (QP). Meanwhile, temporal and spatial windows are designed to select data points for the update of above two R-Q models at frame level and basic unit (BU) level respectively. Furthermore, a linear distortion quantization (D-Q) model is formulated. Combining this D-Q model with the linear R-Q model, a simple close-form solution is developed to derive optimal QP for achieving optimal bit allocation at BU level. At last, based on above three ideas, a rate control scheme is proposed for H.264/SVC base layer. Compared with the rate control scheme JVT-G012 which is adopted by the H.264/SVC base layer, the proposed rate control scheme could increase the average luminance PSNR of reconstructed video by up to 0.58 dB.

A 300bps speech coder based on multi-frame structure and multi-mode matrix quantization is presented. The multi-frame structure consisting of six frames is adopted to reduce the algorithm delay. The parameter matrices are classified into different modes based on the voicing vector information of superframe. To improve speech quality, a dynamic bit allocation scheme is developed. Experimental results show that the speech quality of the proposed vocoder is intelligible with good naturalness.

A 450bps speech coder based on multi-frame structure and multi-mode matrix quantization is presented. The multi- frame structure consisting of four frames is adopted to reduce the algorithm delay. The parameter matrices are classified into different modes based on the voicing vector information of superframe. To improve speech quality, a dynamic bit allocation scheme is developed. Experimental results show that the speech quality of the proposed vocoder is intelligible with good naturalness.

In this paper, the Bilateral Two-dimensional Principle Component Analysis(B2DPCA) and Bilateral Two-dimensional Linear Discriminant Analysis (B2DLDA) are proposed to extract face feature by directly projecting the image matrix. Experimental results on the ORL and PIE face database are performed to test and evaluate the proposed algorithm. The results show that the LDA-based methods outperforms the PCA-based methods, and the two-dimension method outperforms the traditional one-dimensional methods. As opposed to one-dimensional methods, the two-dimensional methods directly extract the proper features from image matrices, while overleaping the process of turning image matrices into vectors, avoid the loss of some structural information residing in original 2D images.

In this paper, we propose a novel algorithm for face feature extraction, namely the cascade two-dimensional locality preserving projections (C2DLPP), which directly extracts the proper features from image matrices based on locality preserving criterion. Experiments on ORL and PIE face database are performed to test and evaluate the proposed algorithm. The results demonstrate the effectiveness of proposed algorithm.

In this paper, we propose a novel algorithm for face feature extraction, namely the bilateral two-dimensional principal component analysis (B2DPCA), which directly extracts the proper features from image matrices. As opposed to PCA, 2DPCA is based on 2D image matrices rather than 1D vector so the image matrix does not need to be transformed into a vector prior to feature extraction. Experiments on ORL and Yale face database are performed to test and evaluate the proposed algorithm. The results demonstrate the effectiveness of proposed algorithm.

LFSR sequence has a good pseudorandom and periodic characteristic. This paper researches the properties of fifth-order fibonacci-lucas sequence, and based on it 5F-L public-key cryptosystem and digital signature scheme are proposed. The fast computational algorithm for evaluating the term of fifth-order fibonacci-lucas sequence is given, and the security and the performance are also discussed. At last, 5F-L algorithms are implemented, tested and analyzed. The result shows that it is very robust and can resist lots of common attacks.

This electronic Propagation is the most important spatial characteristics of bioelectrical signals, whereas time delay estimation (TDE) is the most effective quantitative method in the spatial propagation feature extraction of biomedical signals. A new robust TDE algorithm using the variable step-size affine projection ( RVSAPA) and its application experiments are presented in this paper. Due to it is free of the assumption that the biomedical signals and noises are certain known stochastic distribution, the algorithm is more robust than many traditional analysis methods. Simulation and real application experiments are conducted to precise estimate the evoked potentials (EP) latency changes under background electroencephalograph (EEG) noises and analyze the propagation velocities of gastric myoelectrical activity (GMA) during different phases of the migrating myoelectrical complex (MMC).

A novel face feature extraction method based on Bilateral Two-dimensional Principal Component Analysis (B2DPCA) and Kernel Discriminant Analysis (KDA) was presented in this paper. In this method, B2DPCA method directly extracts the proper features from image matrices at first, then the KDA was performed on the features to enhance discriminant power. As opposed to PCA, B2DPCA is based on 2D image matrices rather than 1D vector so the image matrix does not need to be transformed into a vector prior to feature extraction. Experiments on ORL and Yale face database are performed to test and evaluate the proposed algorithm. The results demonstrate the effectiveness of proposed algorithm

In this paper, an adaptive speaker identification method combined with the human behavioral trait based on Gaussian mixture model (GMM) is constructed. The method can automatically select different length of speech for different speakers in identification process according to the feedback probability estimation, so it can guarantee identification accuracy without reducing, and to reduce the identification time significantly. This method reduces the identification time through two aspects: speaker acoustic features extraction and recognition probability estimation. Experiment results show that for 50 speakers identification, the identification time can reduce 4 times and the identification accuracy can reach 97%. Especially, the new method is fit for large population of speaker identification. From the analysis of this text, we can see that identification time will be reduced more substantially.

In this paper, we propose a novel weighted complete linear discriminant analysis (WCLDA) method for feature extraction and recognition. The WCLDA first introduces a weighting function to restrain the dominant role of the classes with larger distance and then searches the optimal discriminant vectors under the conjugative orthogonal constrains in the null space of the within-class scatter matrix and its conjugative orthogonal complement space, respectively. As a result, the proposed technique derives the optimal and lossless discriminative information. Experiments on ORL and Yale face database are performed to test and evaluate the proposed algorithm. The results demonstrate the effectiveness of WCLDA.

An efficient multi-mode predictive multistage matrix quantization is proposed for very low bit rate quantization of line spectral frequencies (LSF) in LPC based speech coder. One novel aspect of the proposed technique is the mode based predictor and quantizer. Another novel aspect is the optimization of the predictive coefficients considering the mode transition between successive superframes and the connection between predictor and quantizer. Experimental results show that the proposed method yields higher quality LSF quantizer at very low bit rate than those designed with several recently proposed methods.

A* high Q-factor circular-section solenoid-type inductor is designed and fabricated through micro electro mechanical system (MEMS) technology. The radius of the circular-section is 100μm. Ansoft HFSS is used to design and optimize the structure parameters of the inductor. The stable inductance of 10nH and maximum Q-factor of 46 is gained at the N of 10, wire width w of 10μm, space between wires d of 15μm and the self-resonance frequency of the inductor is above 10GHz. A novel fabrication method—flip chip bonding is proposed to bond the two parts of the inductor, and the process is compatible with CMOS process.Copyright © 2008 by ASME

In this paper, we investigate a new blind source separation algorithm for TITO (two-input and two-output) channel system. Considering the case of the noisy instantaneous linear mixture of source signals, we form the matrix pair with the two 2×2 dimension symmetric, positive definite matrices via two covariance matrices. We apply a set of transforms such as the Cholesky factorization and SVD (singular value decomposition) to this formed matrix pair. And a unitary matrix is then obtained, which is an accurate diagonalizer of each matrix of this pair. Compared with the JADE algorithm and the SOBI algorithm, some numerical results show the better performance of the new algorithm.

With ICA to realize the blind separation from mixed acoustic targets, an identification method based on SVM is proposed through extracting LPC feature. SVM is employed to compute the output score and k -means algorithm is used as cluster LPC coefficients. Finally targets are identified by hybrid model. Simulation indicates that this method is effective in mixed acoustic targets identification system.

Based on Takens theory, time delay method is used to reconstruct the phase space of speech signal in this paper. In hyper dimensional phase space, similar sequence repeatability (RPT) and the entropy information of speech are calculated, and they are applied to speech recognition. The result proves that speech signal shows some geometric property between the stochastic Gauss noise and deterministic Lorenz attractor. For speech phonemes, the RPT parameters of same kind show some similarity, and of different kind show some different characteristics as well. The method proposed in this paper provides a new way for the non-linear analysis of speech recognition.