超图拉普拉斯稀疏编码在图像识别中的应用

APPLICATION OF HYPERGRAPH LAPLACIAN SPARSE CODING IN IMAGE RECOGNITION

稀疏编码已经成为一种有效的降维方法。由于编码字典的超完备性、特征之间的局部邻接信息和相似度在编码过程中丢失而降低了稀疏编码的识别率。为了保护特征之间的距离关系和相似信息,提出一种超图稀疏编码框架。这种结构融合相似度权重进入稀疏编码计算过程中,同时结合超图理论,对稀疏编码方法进行改进,增强了稀疏编码的鲁棒性。最后,在Caltech及Scene两大场景数据库上的实验验证了所提方法的有效性。

Sparse coding has been an effective dimensionality reduction method. Due to the over-completeness of the encoding dictionary, the losses of local adjacency information between features and similarity during the encoding process lead to the reduction of the recognition rate of sparse coding. In order to preserve distance relationship between the features and similarity information, we propose a hypergraph sparse coding framework, in this structure the similarity weight is integrated into sparse coding calculation. Meanwhile, hypergraph theory is combined to improve the sparse coding in enhancing its robustness. In end of the paper, the effectiveness of the proposed method is verified through experiments on two scene databases of Caltech and Scene.