摘要：

针对复杂背景下的图像分类问题,结合稀疏编码和邻域保持嵌入算法,提出一种基于邻域保持嵌入规则的稀疏编码算法。在传统稀疏编码问题的目标函数中加入特征编码的局部邻域嵌入正则化项,通过最小化每个特征的编码与其近邻点的特征编码线性组合的误差,使得相似的特征在编码后仍然相似,保留特征的局部邻域结构。采用近似编码的方法降低计算复杂度,实验结果表明,在多个图像数据库上进行图像分类,与已有的稀疏编码方法相比,该算法具有较高的分类精度。

关键词: 图像分类, 稀疏编码, 邻域保持嵌入, 局部邻域结构, 近似编码

Abstract:

Aiming at the problem of image classification with a complex background,this paper proposes a new image algorithm based on Neighborhood Preserving Embedding regularization Sparse Coding algorithm(NPESC).Comparing with traditional sparse coding,it adds the Neighborhood Preserving Embedding(NPE) regularization into the objective function of sparse coding,so that the sparse codes of similar features remain similar by minimizing the deviation between the sparse code of every feature and its neighbors.Thus,this algorithm preserving the local neighborhood structure of the fertures.In addition,approximate coding is used.The result of applying this algorithm to image classification on several datasets,shows that the proposed to reduce computational complexity algorithm can achieve higher accuracy when compared with exissting sparse coding methods.

Key words: image classification, Sparse Coding(SC), Neighborhood Preserving Embedding(NPE), local neighborhood structure, approximate coding

中图分类号: 

• TP391