Introduction to random ferns

TLDCodeThis method is used for machine learning. However, the TLD authors use random forest in their papers. However, random forest and random ferns are all similar.

Recommended link for better English. This link was also provided by a user who previously communicated with TLD. Thank you.

I just want to read the Chinese comrades and follow me to continue to chew on the bones. Although I understand this method, my language expression skills still need to be enhanced. So I found an article about my alma mater on the Internet.ArticleThe following is a basic Excerpt from Dr. Chen ice's paper.

The ferns method uses CI, I = 1, 2,... for sample classification ,... H indicates, set FJ, j = 1, 2 ,... N is the binary feature set of the input sample, and the category of the sample is

Where c Represents a random variable of the class

If the prior probability P (c) is an even distribution, and the (2) denominator is irrelevant to the category, the (1) is simplified

The value of the binary feature FJ here depends on the gray scale idj1 and idj2 of dj1 and dj2 randomly generated during Classifier Training in the sample.

Because FJ is simple in form, it requires enough features to ensure classification results, but the storage required for solving the problem is H * 2 ^ n Bytes. To reduce the storage capacity and ensure sufficient relevance between FJ, and assuming that the binary features of different groups are independent of each other, the binary features in the group are correlated, these groups are defined as ferns feature. Then the conditional probability of formula (3) is approximately

(6) the storage space is only M * 2 ^ s.

these simple formulas are well understood. I hate to use many formulas when introducing things, but these formulas are really simple, if you have read these pictures and do not understand them deeply, you can go to the English link and look at those pictures. It will be easy to handle them. Finally, it is the convention to attach the ferns code, I did not write it, but I still tried to propose a person's work. It's nice to have a foreigner. I dare to share my work with you. Http://cvlab.epfl.ch/software/ferns/index.php. Unfortunately, this code is compiled in the Linux environment, but it is certainly hard for you ~