In this paper,an iterative regularized super resolution (SR) algorithm considering non-Gaussian noise is proposed.Based on the assumption of a generalized Gaussian distribution for the contaminating noise,an lp norm i...In this paper,an iterative regularized super resolution (SR) algorithm considering non-Gaussian noise is proposed.Based on the assumption of a generalized Gaussian distribution for the contaminating noise,an lp norm is adopted to measure the data fidelity term in the cost function.In the meantime,a regularization functional defined in terms of the desired high resolution (HR) image is employed,which allows for the simultaneous determination of its value and the partly reconstructed image at each iteration step.The convergence is thoroughly studied.Simulation results show the effectiveness of the proposed algorithm as well as its superiority to conventional SR methods.展开更多
The non-elementary integrals involving elementary exponential, hyperbolic and trigonometric functions, <img src="Edit_699140d3-f569-463e-b835-7ccdab822717.png" width="290" height="22" ...The non-elementary integrals involving elementary exponential, hyperbolic and trigonometric functions, <img src="Edit_699140d3-f569-463e-b835-7ccdab822717.png" width="290" height="22" alt="" /><img src="Edit_bdd10470-9b63-4b2d-9cec-636969547ca5.png" width="90" height="22" alt="" /><span style="white-space:normal;">and <img src="Edit_e9cd6876-e2b8-45cf-ba17-391f054679b4.png" width="90" height="21" alt="" /></span>where <span style="white-space:nowrap;"><em>α</em>,<span style="white-space:nowrap;"><em>η</em></span><em></em></span> and <span style="white-space:nowrap;"><em>β</em></span> are real or complex constants are evaluated in terms of the confluent hypergeometric function <sub>1</sub><em>F</em><sub>1</sub> and the hypergeometric function <sub>1</sub><em>F</em><sub>2</sub>. The hyperbolic and Euler identities are used to derive some identities involving exponential, hyperbolic, trigonometric functions and the hypergeometric functions <sub style="white-space:normal;">1</sub><em style="white-space:normal;">F</em><sub style="white-space:normal;">1</sub> and <sub style="white-space:normal;">1</sub><em style="white-space:normal;">F</em><sub style="white-space:normal;">2</sub>. Having evaluated, these non-elementary integrals, some new probability measures generalizing the gamma-type and Gaussian distributions are also obtained. The obtained generalized probability distributions may, for example, allow to perform better statistical tests than those already known (e.g. chi-square (<span style="white-space:nowrap;"><em>x</em><sup>2</sup></span>) statistical tests and other statistical tests constructed based on the central limit theorem (CLT)), while avoiding the use of computational approximations (or methods) which are in general expensive and associated with numerical errors.展开更多
The problem of adaptive radar detection in compound-Gaussian clutter without secondary data is considered in this paper.In most practical applications,the number of training data is limited.To overcome the lack of tra...The problem of adaptive radar detection in compound-Gaussian clutter without secondary data is considered in this paper.In most practical applications,the number of training data is limited.To overcome the lack of training data,an autoregressive（AR）-process-based covariance matrix estimator is proposed.Then,with the estimated covariance matrix the one-step generalized likelihood ratio test（GLRT） detector is designed without training data.Finally,detection performance of our proposed detector is assessed.展开更多
贝叶斯概率矩阵分解方法因较高的预测准确度和良好的可扩展性,常用于个性化推荐系统,但其推荐精度会受初始评分矩阵稀疏特性的影响.提出一种基于广义高斯分布的贝叶斯概率矩阵分解方法GBPMF(generalized Gaussian distribution Bayesian...贝叶斯概率矩阵分解方法因较高的预测准确度和良好的可扩展性,常用于个性化推荐系统,但其推荐精度会受初始评分矩阵稀疏特性的影响.提出一种基于广义高斯分布的贝叶斯概率矩阵分解方法GBPMF(generalized Gaussian distribution Bayesian PMF),采用广义高斯分布作为先验分布,通过机器学习自动选择最优的模型参数,并基于Gibbs采样进行高效训练,从而有效缓解矩阵的稀疏性,减小预测误差.同时考虑到评分时差因素对预测过程的影响,在采样算法中添加时间因子,进一步对方法进行优化,提高预测精度.实验结果表明:GBPMF方法及其优化方法 GBPMF-T对非稀疏矩阵和稀疏矩阵均具有较高的精度,后者精度更高.当矩阵非常稀疏时,传统贝叶斯概率矩阵分解方法的精度急剧降低,而该方法则具有较好的稳定性.展开更多
基金National Natural Science Foundations of China(No.60705012,No.60802025)
文摘In this paper,an iterative regularized super resolution (SR) algorithm considering non-Gaussian noise is proposed.Based on the assumption of a generalized Gaussian distribution for the contaminating noise,an lp norm is adopted to measure the data fidelity term in the cost function.In the meantime,a regularization functional defined in terms of the desired high resolution (HR) image is employed,which allows for the simultaneous determination of its value and the partly reconstructed image at each iteration step.The convergence is thoroughly studied.Simulation results show the effectiveness of the proposed algorithm as well as its superiority to conventional SR methods.
文摘The non-elementary integrals involving elementary exponential, hyperbolic and trigonometric functions, <img src="Edit_699140d3-f569-463e-b835-7ccdab822717.png" width="290" height="22" alt="" /><img src="Edit_bdd10470-9b63-4b2d-9cec-636969547ca5.png" width="90" height="22" alt="" /><span style="white-space:normal;">and <img src="Edit_e9cd6876-e2b8-45cf-ba17-391f054679b4.png" width="90" height="21" alt="" /></span>where <span style="white-space:nowrap;"><em>α</em>,<span style="white-space:nowrap;"><em>η</em></span><em></em></span> and <span style="white-space:nowrap;"><em>β</em></span> are real or complex constants are evaluated in terms of the confluent hypergeometric function <sub>1</sub><em>F</em><sub>1</sub> and the hypergeometric function <sub>1</sub><em>F</em><sub>2</sub>. The hyperbolic and Euler identities are used to derive some identities involving exponential, hyperbolic, trigonometric functions and the hypergeometric functions <sub style="white-space:normal;">1</sub><em style="white-space:normal;">F</em><sub style="white-space:normal;">1</sub> and <sub style="white-space:normal;">1</sub><em style="white-space:normal;">F</em><sub style="white-space:normal;">2</sub>. Having evaluated, these non-elementary integrals, some new probability measures generalizing the gamma-type and Gaussian distributions are also obtained. The obtained generalized probability distributions may, for example, allow to perform better statistical tests than those already known (e.g. chi-square (<span style="white-space:nowrap;"><em>x</em><sup>2</sup></span>) statistical tests and other statistical tests constructed based on the central limit theorem (CLT)), while avoiding the use of computational approximations (or methods) which are in general expensive and associated with numerical errors.
基金supported by the Fundamental Research Funds for the Central Universities under Grant No. E022050205
文摘The problem of adaptive radar detection in compound-Gaussian clutter without secondary data is considered in this paper.In most practical applications,the number of training data is limited.To overcome the lack of training data,an autoregressive（AR）-process-based covariance matrix estimator is proposed.Then,with the estimated covariance matrix the one-step generalized likelihood ratio test（GLRT） detector is designed without training data.Finally,detection performance of our proposed detector is assessed.
