###### tags: Paper Reading
# Search-engine-for-large-coleection-of-image
   This paper is talking about how to build a large face Image search engine.
## Introduction/Motivation
    Nowadays, lots of image take by users are released to social network. While the number of image growing, how to maintain and search a image become a problem, especially in searching image with human face. This paper announce a large collections of image with face. We can learn some tricks from the construction of process.
## Database
    Although the detail of creating a search engine is also wonderful part of the paper , but I will focus on the preprocess of the image on this paper. If you want details of database, please go to read **section 3**.
## Classification trick
&#160; &#160; The preprocess of image of the engine is a most important tick that we should learn. This paper create <span class="blue">**rich set of local feature**</span> options for classifiers instead of send <span class="red">**whole pixels of the face**</span> into classifier. This is a very novel point of view for me.
&#160; &#160; The idea of this policy is that while we don’t know which feature is a good choice of the attribute. Why not we split the face into some local features, and let machine learn which local feature is important?
&#160; &#160; Taking the advantage of adaboost, we can realize previous idea. The paper also provide some options for classifier like
* face regions,
* types of pixel data,
* normalizations
* aggregations.
&#160; &#160;
## Conclusion
&#160; &#160; After training the classifier we can know which feature is useful for the attribute that we call it explainable AI.
<style>
.blue {
color: blue;
}
.red {
color: red;
}
</style>