基于ICA的谱聚类方法在特征捆绑中的应用

Spectral clustering based on ICA for feature binding

为探究视觉系统处理图像颜色和形状时的特征捆绑问题,提出了基于独立成分分析的谱聚类方法。利用独立成分分析对任务态下的fMRI数据进行成分的提取,利用谱聚类算法对成分和任务之间的相关系数进行谱聚类分析,得到一种基于独立成分的谱聚类方法。将该算法应用于颜色和形状的特征捆绑任务中,结合提出的认知减法,得出了任务态下参与特征捆绑的主要激活脑区,并对任务态下认知颜色和形状的主要脑区进行了分析研究,为建立视觉特征捆绑的认知模型提供理论基础,表明了该方法适用于多任务的fMRI数据分析。

Spectral clustering method based on independent component analysis is proposed to explore feature binding of color and shape in visual system. First independent component is extracted from task fMRI data. Then correlation coefficient between the compositions and tasks is calculated. Finally correlation coefficient matrix is clustered. Spectral clustering algorithm based on in dependent component is acquired. The algorithm is applied to the feature binding of color and shape combined with the proposed cognition subtraction. The conclusion is drawn that activate brain regions involved in feature binding task and the main brain re gions which cognize color and shape are analyzed. Theoretical basis is provided for the establishment of the cognitive model of visual feature binding. The results prove the algorithm is applicable to multi-task fMRI data.