提出一种新的对象分类方法，该方法将抠像技术应用到图像检索领域。给前景层图像块较高的权值，背景层图像块较低的权值，并对图像块按照权值大小排序，图像间内容相似度比较时，首先计算对应权值的图像块间的距离，再将所有块间距离加权累加作为图像间距离。新方法的有效性在牛津大学花卉图像集上得到验证，实验结果也表明，相对于传统的计算对应位置的图像块间距离的方法，新方法的检索准确度有明显的优势。

This paper presents a novel method for recognizing object categories. In this method, matting technique is applied to estimate foreground distribution information, which provides weights for image blocks. Blocks describing foreground are given higher weights, while blocks describing background are given lower weights, then all the blocks are sorted according to the assigned weight.We also conducted imagery comparison study. First of all, distances between blocks with corresponding weights are evaluated, and then the similarity between images can be obtained through adding all the distances between blocks by weights. Experiments on challenging object categorization image sets demonstrate the effectiveness of this method.