Sparse coding can achieve good performance in some computer vision problems.However,past sparse coding was implemented in the original feature space.Kernel method can acquire high dimensional nonlinear mapping characteristics.Inspired by it,the Laplacian sparse coding(LSc)is extended,and the kernel Laplacian sparse coding(KLSc)is proposed.It can reduce the feature quantization error and enhance the sparse coding performance.Experimental results of three standard datasets show that the proposed image classif...
【基金】
国家自然科学基金资助项目(61371157)
【更新日期】
2015-04-09
【分类号】
TP391.41
【正文快照】
0引言近年来,词包(bag of words,BoW)模型[1]由于在尺度、平移和旋转等方面具有很强的鲁棒性而备受关注.研究者提出很多扩展BoW模型,例如,模拟码字或描述子同现的生成算法[2];替代标准无监督聚类算法的字典学习方法[3];模拟局部特征空间分布的空间金字塔匹配核[4]等.在这些扩展