基于多示例学习的图像分类算法

Image Categorization by Multi-instance Learning

基于内容的图像分类计数通常基于图像的单一特征进行处理,而图像中包含的内容不止一个,单一的特征不足以充分描述图像,多实例学习方法由于其特殊性可以很好地解决这个难题。文中针对基于多示例学习的图像分类问题提出了一种新的多示例学习算法DD-TSVM。该方法以图像作为包,图像中的区域作为包中示例。算法首先采用多样性密度算法寻找示例集的局部最大值以构建投影空间并将包映射为投影空间中的一个点;然后利用直推式支持向量机作为学习算法训练学习得到分类器。该算法有效地利用了未标记样本,基于Corel图像数据库的实验结果表明,DD-TSVM具有良好的性能。

There have been great achievements in CBIC,and only the one single feature is generally used in the methods. Since there is more than one object in an image,it is not enough to use one feature to describe the image. The method of multi-instance learning can deal with the above problem. For the problem of multi-instance-based image categorization, a creative multi-instance learning algorithm named DD-TSVM has been proposed. This algorithm regards the image as a bag, and the region of image as an instance in the bag. First, a local maximum set has been found by diversity density algorithm to construct a projection space and transform each bag into a point in the projection space. Then, using transductive support vector machine is to get the classifier. The proposed algorithm effectively takes ad- vantage of the unlabelled samples. The experimental results on Corel dataset show that DD-TSVM has good performance.