Multi-feature image retrieval based on Tri-training
Chen Xiu-xi · Jisuanji yingyong yanjiu · 2014
To effectively use multi-features in images for image retrieval,this paper used Tri-training to fuse the color feature,texture feature and shape feature. First,it extracted the three-dimensional quantized color histogram,steerable pyramid binary image projection and affine invariant regions to represent the color feature,texture feature and shape feature of the image.Then,the matching results of these three features served as the input of the Tri-training classifier for its training and testing.Experiments show that by the fusion of these features,this method is very effective.