多类SVM在图像艺术属性分类中的应用研究

An application of multi-class SVM in the classification of artistic attributes of images

针对当前图像分类研究中,依据图像艺术风格属性进行分类的算法尚不多见的情况,实现了一种基于艺术属性的图像自动分类系统,其中主要涉及摄影作品、国画、水彩画、素描、油画等几种典型艺术风格的图像.系统采用支持向量机(SVM)作为分类器,运用分等级的分类方法,提出了一种针对艺术属性图像分类的特定SVM二叉树多类分类算法;而后通过对各类图像艺术风格特征的分析,分别提取了有代表性的、区分度好且易于计算的特征;最后针对各级分类特性和分类器总体特性进行了实验分析,实验结果表明,系统具有良好的分类性能.

In image classification, few current classification algorithms classify images by their artistic attributes. An automatic image classification system based on artistic attributes was developed for classifying images in typical artistic styles such as photographs, Chinese paintings, watercolors, sketches, oil paintings, and so on. The system employed a support vector machine （SVM） as a classifier. Using a classification method at various levels, an SVM binary tree multi-class classification algorithm for image classification with respect to different artistic attributes was proposed. By analyzing the images with respect to the different artistic styles, some easily computed representative characteristics with good discriminability were extracted at each classification level. Experiments on a variety of characteristics at various levels and the total characteristics of classifiers were designed to evaluate the proposed classifier. Experimental results showed that the system has good classification performance.