模式分类中的特征融合方法

特征融合方法是模式识别领域的一种重要方法.计算机视觉领域的图像识别问题作为一种特殊的模式分类问题,仍然存在很多挑战.特征融合方法能够综合利用多种图像特征,实现多特征的优势互补,获得更加鲁棒和准确的识别结果.笔者基于信息融合理论分析了特征融合方法的原理,介绍了特征融合方法的研究现状,讨论了特征融合与3类主流基础理论相结合的方法,其中基于贝叶斯理论的特征融合算法可以实现多特征的融合决策,基于稀疏表示理论的特征融合算法能够得到多特征的联合稀疏表示,基于深度学习理论的特征融合算法能够强化深度神经网络模型的特征学习过程.

Feature Fusion Methods in Pattern Classification

Feature fusion is an important method in pattern recognition.Image recognition problem in computer vision can be known as a special pattern classification problem,and it still have many challenges.To solve the problem,feature fusion method is able to use multi-feature of image,complement each other's advantages and get more robust and accurate results.Based on information fusion theory,analyze the theory of feature fusion and introduce the development of feature fusion methods.Besides,discuss three methods of feature fusion combining with other basic theories.Bayesian decision theory based feature fusion algorithm implements fusion decision of multi-feature.Sparse representation based feature fusion algorithm can get the joint sparse representation of multi-feature.Deep learning based feature fusion algorithm intensifies feature learning process of deep neural network model.