基于集成分类算法的自动图像标注

摘要

Automatic image annotation is an important but highly challenging problem in semantic-based image retrieval. In this paper, we formulate image annotation as a supervisee! learning image classification problem under the region-based image annotation framework. In region-based image annotation, keywords arc usually associated with individual regions in the training data set. This paper presents a novel ensemble fast random rarest algorithm (FRF), which can classify a large number of training data effectively by bootstrap aggregation (Bagging) algorithm building multiple tree component classifier. The final result is obtained by component classifier voting. The proposed FRF algorithm is experimented on image annotation Corel data sets. Compared to classical algorithms, the FRF accelerates the operation speed of the algorithm, and the classification accuracy remains robust. It has a good application in automatic image annotation system.%基于语义的图像检索技术中,按照图像的语义进行自动标注是一个具有挑战性的工作.本文把图像的自动标注过程转化为图像分类的过程,通过有监督学习对每个图像区域分类并得到相应关键字,实现标注.采用一种快速随机森林(Fast random forest,FRF)集成分类算法,它可以对大量的训练数据进行有效的分类和标注.在基于Corel数据集的实验中,相比经典算法,FRF改善了运算速度,并且分类精度保持稳定.在图像标注方面有很好的应用.