补丁校准框架下的最大模糊边界投影

Maximum Fuzzy Marginal Projection via Patch Alignment Framework

补丁校准是一类有效的维数简约框架。基于补丁校准框架,提出一种最大模糊边界投影算法。该算法引入模糊集理论,从样本的相似度出发,利用非负最小二乘法获取相似近邻,进而构造相似隶属度矩阵,依据相似隶属度矩阵重新定义了模糊边界补丁中心和模糊相似权重。模糊边界补丁中心能很好地降低(或消除)重叠(离群)样本对于特征提取的影响;而模糊相似权重明确了该样本对特征提取所做的贡献。在补丁校准框架下,同类样本间由光照、表情等变化所引起的差异能得到有效的压制,同时不同类样本间距离得以增大,有助于分类性能的提高。在UCI Wine、Yale和Yale-B数据库上的实验验证了所提方法的有效性。

Patch alignment（PA）framework provides us a useful way to obtain the explicit mapping for dimensionality reduction.Under the PA framework,we proposed a fuzzy maximum marginal projection for dimensionality reduction.In our paper,the fuzzy set theory was introduced in the design of new method.The similar neighbors obtained by the nonnegative least squares method were used to construct the similar membership degree matrix.Based on the similar membership degree matrix,we redefined the fuzzy weight and the fuzzy marginal patch means.The fuzzy weight can reduce the influence caused by the overlap and outliers to some extent.The fuzzy marginal patch means specify the contributions of each sample to the classification.Under the PA framework,the difference among intra-class samples caused by the variety of the illumination can be degraded.And the distances between different categories are enlarged in the transformed space.The experimental results on the UCI Wine,Yale and Yale-B databases demonstrate the effectiveness of the new methods,especially in dealing the changing illumination on images.