Learning from noisy labels (LNL) aims to train high-performance deep models using noisy datasets. Meta learning based label correction methods have demonstrated remarkable performance in LNL by designing various meta ...
Efficient linear separation algorithms are important for pattern classification applications. In this paper, an algorithm is developed to solve linear separation problems in n-dimensional space. Its convergence featur...
详细信息
Efficient linear separation algorithms are important for pattern classification applications. In this paper, an algorithm is developed to solve linear separation problems in n-dimensional space. Its convergence feature is proved. The proposed algorithm is proved to converge to a correct solution whenever the two sets are separable. The complexity of the proposed algorithm is analyzed, and experiments on both randomly generated examples and real application problems were carried out. While analysis shows that its time complexity is lower than SVM that needs computations for quadratic programming optimization, experiment results show that the developed algorithm is more efficient than the least-mean-square (LMS), and the Perceptron.
Optimal thresholds selection for multiwavelet denoising using multivariate shrinkage is considered. Unlike other risk estimators which are derived for a single threshold, a new risk estimator is proposed for each deco...
详细信息
Optimal thresholds selection for multiwavelet denoising using multivariate shrinkage is considered. Unlike other risk estimators which are derived for a single threshold, a new risk estimator is proposed for each decomposition level based on the principle of Stein's unbiased risk estimator.
The purpose of this paper is to investigate the accuracy of PESQ and 3SQM in predicting voice quality over live 3G networks. We set up an Asterisk-based testbed to measure voice quality over 3G mobile networks. 192 ex...
详细信息
The contourlet transform has emerged recently as an efficient directional multiresolution image decomposition. In order to enhance its distinguishing power for texture images, the high frequency band is split by half ...
详细信息
ISBN:
(纸本)0980326702
The contourlet transform has emerged recently as an efficient directional multiresolution image decomposition. In order to enhance its distinguishing power for texture images, the high frequency band is split by half and results in the multiscale directional filter bank (MDFB). However, this raises the computational complexity considerably because an extra set of scale and directional decomposition is required to perform on the full image size. In this paper, we develop a new framework for the MDFB, in which directional decomposition is performed prior to the scale decomposition in the first two scales. Experiments show that 30.6%-33.2% saving in computational time can be achieved. Meanwhile, it has the same flexibility in frequency partitioning and similar texture retrieval performance as the original algorithm of the MDFB.
Z-curve features are one of the popular features used in DNA sequence classification. Here, we studied the Z-curve features from a signalprocessing point of view. In particular, the Z-curve features are re-interprete...
详细信息
ISBN:
(纸本)0980326702
Z-curve features are one of the popular features used in DNA sequence classification. Here, we studied the Z-curve features from a signalprocessing point of view. In particular, the Z-curve features are re-interpreted through a spectral formulation. Our analysis showed that there are significant differences in the spectral interpretation between the Z-curve formulation and the FFT (Fast Fourier Transform) approach. From the spectral formulation of the Z-curve approach, we obtained three modified sequences that characterize different biological properties which are useful for coding region prediction. Spectral analysis on the modified sequences showed a much more prominent three-periodicity property in coding regions than using the FFT approach. Our experiments indicated that for long sequences, prominent peaks at 2II/3 are observed at coding regions. For short sequences, peaks can still be observed at coding regions. We also obtained good classification performance using the spectral features derived from the three modified sequences.
An efficient realization of symmetric orthogonal discrete multiwavelet transform using simple scaling and rotation matrix factorization is proposed in this paper. Simulation results show that, by factorizing the polyp...
详细信息
Restoring an image from its convolution with an unknown blur function is a well-known ill-posed problem in image processing. The generalized cross validation (GCV) approach was proposed to solve the problem and it has...
详细信息
In this paper, an efficient automatic human face recognition system is proposed. Fractal dimension is an efficient representation of texture which is used to locate the eyes in a human face. We propose a modified appr...
详细信息
A new temporally scalable video coding algorithm based on the interpolating wavelet transform is proposed. With the proposed approach, the input video frames are first applied to an interpolating wavelet transform whi...
详细信息
暂无评论