【摘要】 为了有效地解决多示例图像分类问题,基于稀疏表示提出了一种新的多示例图像分类方法。该方法将图像看作多示例包,图像中的区域作为包中示例,利用示例嵌入策略计算包特征;然后将待分类图像包特征表示为训练图像包特征集上的稀疏线性组合,利用1优化方法求得稀疏解;最后根据稀疏系数提出一个为待分类图像预测标记的方法。在Corel数据集上的实验结果表明,与其他方法相比,所提方法具有更高的分类精度。更多还原

【Abstract】 In order to effectively solve the problem of multi-instance image classification,a novel classification method of multi-instance image was proposed which is based on sparse representation.The whole image is regarded as a bag and each region as an instance of that bag.The image bag feature is computed based on instance embedded strategy.Next,the test image bag feature is regarded as sparse linear combination of training image bag feature set and the sparse solution can be obtained by 1optimization method.Finally,sparse coefficients are utilized to predict the label of the test image.Experimental results on the Corel image data show that the proposed method is superior to the state-of-art methods in terms of classification accuracy.更多还原