提出了一种基于全变分正则与L_(2,1)范数的视频去雨张量模型用于解决雨线遮挡问题。首先,对雨线成分与视频背景先验信息进行预处理,获取相应正则化条件的构建依据以增强各部分稀疏性,便于促进雨线分离。其次,考虑到视频图像存在不规则动态对象,引入全变分正则项来抑制背景强度变化,缓解雨线的误判现象。采用交替方向乘子法(alternating direction method of multipliers,ADMM)可以有效地对所提出的张量模型进行求解,并在合成数据与真实数据集上开展大量实验。结果表明,所提方法在动态背景情况下有效去除视频图像雨线的同时,保留了更多背景细节信息。与相关先进方法相比,所提方法在峰值信噪比、结构相似性和残差三种综合性能量化指标上均具有较大的优势。
针对Fisher判别法判别同族群体多中心数据准确率低的问题,提出了线性映射模型下的重编码判别分析算法,将Fisher判别法中的降维思想与重编码方法相结合,采用蒙特卡洛法,通过对伪预测数据的划分。以实映射识别率为目标,确定线性判别函数的待定系数和伪预测数据的划分。实证表明,该算法具有较高的识别率和稳定性。Aiming at the problem of low accuracy of Fisher discriminant method in judging the multicenter data of the same population, a recoding discriminant analysis algorithm under linear mapping model is proposed, which combines the idea of dimension reduction in Fisher discriminant method with recoding method, adopts Monte Carlo method, and divides the pseudo-predicted data. Aiming at the recognition rate of real mapping, the undetermined coefficients of linear discriminant function and the division of false prediction data are determined. The empirical results show that the algorithm has high recognition rate and stability.
暂无评论