关键词: 图像分类, 和积网络, 特征提取, 结构学习

Abstract: This paper proposes a new flower image classification method, throughout studying a new sum-product network structure, and combining it with the novel image feature extraction method. The new classification method, which takes the characteristics of natural scene images into consideration, and uses the information of image patches to extract the feature, makes the extracted feature vectors distinguishable, independent and robust. The use of sum-product structure learning algorithm for matrix is propitious to group similar instances into the same cluster and segment distinct variables into different categories. Experimental results show that the method has a better performance in flower image classification.

Key words: image classification, sum-product networks, feature extraction, structure learning