一种多特征融合的场景分类方法

Multi-feature Fusion Method for Scene Classification

针对图像当中的不同对象,各种特征的优势各不相同,彼此之间存在互补现象.因此,提出一种多特征融合的图像场景分类方法.首先分别提取图像的GIST特征、SIFT特征和PHOG特征;然后将SIFT特征进行局部约束线性编码,并基于空间金字塔模型进行最大池化生成稀疏向量表示;接着采用串联的方法将GIST特征、SIFT特征稀疏向量表示和PHOG特征进行特征融合;最后将融合特征与类标签信息一起输入到线性SVM进行分类.多特征融合的图像场景分类方法,充分考虑了各个特征之间的优势以及图像原有特性和单词空间分布,能够有效的达到特征互补.实验结果表明,与其他分类方法相比,该方法具有较好的分类性能.

In view of the different objects in the image,the advantages of various features are different,and there is complementarity between each other,this paper proposed a multi-feature fusion scene classification method. Firstly,the GIST feature,SIFT feature and PHOG feature are extracted respectively; Then,the SIFT features are encoded by the locality-constrained linear coding and max-pooled based on the spatial pyramid model to generate sparse vector representations; Next,The fusion feature of GIST feature,SIFT feature sparse vector and PHOG feature is performed by serial method; Finally,the fusion feature and the class label information are used as input to train linear SVM classifiers. The image scene classification approach based on multiple features fusion takes full account of the advantages of each feature. The original characteristics and word spatial distribution of the image can achieve feature complementation effectively. The experimental results show that the classification accuracy of the proposed method is improved compared with other methods.