一种基于图像簇多核学习模型的图像分类方法

Image Classification Based on Image Cluster Multiple Kernel Learning Model

依靠单一特征核函数不能很好地表示图像语义内容的融合,导致其图像分类算法只具备一般的判别能力。针对上述问题,采用基于图像簇模型的图像分类方法,研究将给定样本图像类别中具有相似形状、纹理或颜色特征的样本图像聚类到一个图像簇中,确保图像分类的准确性;该方法解决了样本图像簇和单个样本目标之间的关系,即样本图像中单个图像之间的差异,使学习模型获得较高的判别能力。经实验验证,图像簇多核学习模型能很好地减弱类内差异性和类间相似性的影响,且得到鲁棒性更强的图像分类模型。

Single feature kernel function can not well representthe fusion of semantic content of image,and the image classification algorithm has only general discriminant ability.To solve these problems,image classification based on image cluster multiple kernel learning model is adopted,introducing in detail the given sample image categories with similar characteristics of shape,texture or color sample image clustering into an image cluster,thus ensuring the accuracy of image classification.The relationship between the cluster of sample images and the target of a single sample is determined,that is,the difference between the single images in the sample images,which brings learning model higher discriminant ability.Experimental verification on the databases Scene15,UIUC Sports and caltech-256 shows that the image set MKL model can effectively weaken the influence of intra-class difference and inter-class similarity,a nd obtain a more robust image classification model.