what is feature representation?
no matter what learning and analytical architecture you adopt, the initial procedure refers to getting the basic feature expressions from the training datasets which has a large scale than the test datasets, and then try the proposed linear or non-linear algorithms to combine the basic featrues to get the possible result near to the final to-be-represent images or other formats objects, and the feedback part could also be set to adjust the parameters of the learning algorithms to repair the gotted result.
静脉思路:
静脉纹路其实是一种比较简单的纹路结构,所以可以引入稀疏编码的思路,设置尽可能多的训练集,通过算法设计训练得到稀疏编码中的基特征,随后对输入的静脉特征进行feature representation,将特征表示中的各基特征对应的系数作为该静脉特征表示,结合图像检索思路进行身份匹配;
另外一种思路是将上述的基特征用基本的直方图分布代替,将直方图组合作为特征表示;