一种基于层次语义图像分类的改进方法

AN IMPROVEMENT METHOD BASED ON HIERARCHICAL SEMANTIC IMAGES CLASSIFICATION

为了解决层次语义图像中分类率低,特别是高层语义图像分类率低的问题,采用两种解决措施。首先引入Fuzzy Support Vector Machine(FSVM)理论,并对FSVM做出改进,消除由Support Vector Machine(SVM)构成的多类分类器中的不可分区域,从而使低层语义图像分类准确率提升,为高层语义分类提供基础。然后再建立底层图像特征与低层语义图像之间的映射关系,对低层语义的图像做高层语义上的关联,最终实现层次化的语义描述结构。实验表明,所提出的方法提高了层次语义图像,特别是高层语义图像分类准确率。

This paper uses two solutions for the problem of low classification rate in hierarchical semantic images, in particular the high-level hierarchical semantic images. Firstly we introduce the theory of fuzzy support vector machine （ FSVM ） and improve it, this eliminates the unclassifiable region of the multi-class classifiers constructed with support vector machine （ SVM ）, therefore the image classification accuracy rate of lower-level semantic images is enhanced ; it provides a basis for the high-level semantic classification. Then, we establish the mapping relationship between the bottom image characteristics and the lower-level semantic images for making the association of high-level semantics for low-level semantic image, and finally achieve the hierarchical semantic description structure. Experimental results show that the presented method can improve the classification accuracy rate of hierarchical semantic images, especially of the high-level hierarchical semantic images.