一种协同演化下的二元邻域判定误分纠错方法  

A Binary Neighbors Discriminative Method in Collaborative Evolution for Error Correction

作　　者：吴克伟[1] 谢昭[1] 高隽[1] 赵碧茹[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《计算机学报》2014年第11期2276-2286,共11页Chinese Journal of Computers

基　　金：国家自然科学基金(60905005;61273237);中央高校基本科研业务费专项资金(2012HGCX0001);国家"九七三"重点基础研究发展规划项目基金(2013CB329604)资助~~

摘　　要：针对图像分类过程中出现误分样本的问题,提出了一种协同演化的计算方法,可有效纠正误分点.在数据空间协同表示的基础上,考虑误分样本周围的支持样本和竞争样本对该类样本的协同作用,将误分样本逐渐拉入正确的决策区域内,实现误分样本的正确分类,达到纠错目的,同时从理论上证明了方法的收敛性.仿真数据集和真实图像集分类的实验验证了协同演化计算策略能够稳定收敛至唯一点,结合高维空间中ISOMAP流形,提出的协同演化模型有较好的纠错能力,并较已有方法在低约简维数上有更好的执行效率和分类性能.Since there are many misclassification samples in image classification process,it results in low classification accuracy.For the purpose of effectively correcting misclassification samples,this paper presents an approach called collaborative evolution computation（CEC）strategy.On the basis of collaborative representation,we take the synergy of support samples and competition samples around misclassification samples into consideration,gradually make misclassification samples pulled into correct decision area,and finally obtain the achievement of making misclassification sample correctly classified.The experiments on simulation data set and real image set provide the validation that a misclassification sample could converge into a unique point by using CEC strategy.In the combination with ISOMAP manifold,we propose collaborative evolution model which is proved to be a stable and convergent model.Compared with state-of-the-art image classification methods,this model achieves better efficiency and classification performance on low dimensions.

关 键 词：协同进化 流形 近邻空间 二元决策 数据优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]