特征级融合方法及其在医学图像方面的应用

FEATURE-LEVEL FUSION AND ITS APPLICATION IN MEDICAL IMAGE

特征级融合是图像处理领域的重要内容之一,从特征变换和特征选择两个角度对特征级融合方法及其在医学图像处理中的应用进行较为全面的总结。以医学图像特征级融合为例介绍其基本流程,将特征级融合方法分为特征变换和特征选择。在特征变换方面,按照方法是否线性可分和是否监督学习两个维度进行划分总结,详细介绍线性变换方法中奇异值分解法和非负矩阵分解法,非线性变换中的人工神经网、支持向量机和模糊集五种方法及其在医学图像方面的应用。总结特征选择方法的整体流程,分别从搜索策略和评价准则的角度对方法进行分类总结,详细介绍粗糙集和遗传算法两种随机搜索算法的改进及其在医学图像方面的应用,并阐述停止条件和结果验证过程。对特征级融合方法的发展方向进行总结和展望。特征级融合算法的总结对其进一步发展具有积极意义。

Feature-level fusion is one of the most important parts in the field of image processing. From the two perspectives of feature transformation and feature selection, this paper summarized the feature-level fusion and its application in medical image processing comprehensively. We introduced the basic process of medical image feature-level fusion, and it was divided into feature transformation and feature selection. In the aspect of feature transformation, it was divided into two dimensions: whether the method was linear separable and whether supervised learning or not. We introduced in detail the singular value decomposition and non-negative matrix decomposition in linear transformation, artificial neural network, support vector machine and fuzzy set methods in non-linear transformation and their applications in medical images. The overall process of feature selection method was summarized. The methods were classified and summarized from the perspective of search strategy and evaluation criteria. We introduced in detail the improvement of two random search algorithms, rough set and genetic algorithm, and their applications in medical images. The stopping conditions and the verification process of the results were also described. We summarized and prospected the development trend of feature-level fusion. The summary of feature-level fusion has positive significance for its further development.