本文提出一种场景分类方法,通过整合局部特征和滤波器特征获得丰富的表征信息,并利用空间金字塔匹配模型提取空间上下文信息.该方法有如下四个特点:(1)通过转换将滤波器很好地嵌入空间金字塔匹配模型中;(2)在滤波器特征转换的过程中,采用降采样和平均操作,在空间密度和空间范围两者之间取得了很好的折衷;(3)将滤波器特征和局部特征组合起来,获得了更强的描述能力;(4)捕获了像素域和调制域的互补信息.同时,在三个数据库上的实验证明了该方法的有效性.
This paper presents an approach to scene classification,which unifies local features and filterbank features to capture rich representation information,and extracts spatial context information using the spatial pyramid matching(SPM) model.The proposed method has four characteristics.First,filterbank features are successfully embedded into the SPM model by a transformation method.Second,in the transform process,downsampling and average pooling are used to achieve good balance between spatial density and spatial extent.Third,filterbank features and local features are combined to represent images for more discriminative power.Fourth,the complementary information is extracted in pixel and modulation domains.Promising experimental results on three datasets demonstrate the effectiveness of the proposed method.