基于多尺度稀疏表示的场景分类

提出一种基于多尺度稀疏表示的场景分类框架。首先从图像中提取多个尺度的局部特征,然后利用稀疏编码为每个尺度的特征单独学习相应的过完备字典;在图像表示阶段,为图像各尺度上的局部特征依据与其对应尺度的字典进行编码,并按照空间金字塔表示方法和特征各维最大汇总(max pooling)对各尺度上的特征编码分别汇总;最后将不同尺度上汇总的特征串接,形成对图像最终描述的全局向量。在三个常用标准场景库上的分类结果表明,提出的算法由于利用了不同尺度特征间的互补关系,与采用单尺度特征的方法相比,性能有了显著提升。

Multi-scale sparse representation based scene categorization

This paper proposed a novel multi-scale sparse representation based framework of scene categorization. Firstly, it extracted local features of multiple scales, then learned dictionaries of each scale by using sparse coding algorithm. In the stage of image representation, this paper proposed to encode each local feature according to dictionary on the same scale. By using spatial pyramid representation and feature max-pooling, it summarized the features on each scale. In the final step, the proposed algorithm concatenated the pooled features of different scales to form the holistic feature. The experimental results on three popular benchmark datasets show that since the proposed method utilizes the complementary relationship of features on different scales, it obtains better performance compared with methods using features on single scale.