Question

关于用PCA做人脸识别

我想问的是:
(1 PCA做人脸识别的机制;
(2 PCA的基础K-L变换到底怎么理解;
希望能给个通俗易懂的回答,资料我这很多,主要我希望有人能让这个刚涉及这方面知识的我有个稍微有点深入的了解.

附加:希望与有这方面研究的朋友一起交流;
邮箱:z123321@163.com

悬赏分先给 100 吧!如果给的回答能让我有更深入的了解,我将在额外加200悬赏分.

Answer 1

PCA is a vector space transform often used to reduce multidimensional data sets to lower dimensions for analysis.

your question 1 is somehow wrong.

I do not know how much you have already know about pattern classification. Generally, we represent an object with a vector, the more attributes it has, the dimensions the vector has. For example, a 25 years old male [25, 0] a 35 years old female [35,1]

However, an object always has a lot of attributes, which makes the dimension of the vectors very high. This fact leads to huge computation and worse result (in most cases). We can understand it like this, of course to represent you alone, you need a lot of attributes, say, 100. But to distinguish you from me, only 10 will be enough. The rest of the 100 numbers then becomes "noise" in our classifier.

This is why we need to do "feature selection", "dimension reduction" etc.

PCA reduces dimension by retaining those characteristics of the data set that contribute most to its variance. You may want to read some paper about SVD to understand it better.

For K-L transform, actually I never heard that before. But I guess it might be another name of SVD.

If you have any question, feel free to contact me (by send messages).