main features of k means clustering

I am working on cbir using k means clustering,my first step is to extract features ,can you say what are all the main features to extract for k means clustering,

i have extracted one colour(HSV),and have clustered ,can tou say other two features to extract,

Answer 1

[Walter Roberson]

Pick any two different functions that take an image as input and produce a scalar or vector as output. The output of those functions will each be "features".

The features that would be recommended would depend a lot on your purpose, and on the characteristics of the images that are to be processed.

For example, if you are processing solar images, then if your purpose is to study the solar corona, you would be wanting to ignore "space", but if your purpose is to study Near Solar comets, you would be wanting to ignore the Sun and pay attention to fine details in "space".

[kash]

I am focussing on satellite Images ,what would be appropiate features

[Walter Roberson]

KEPLER or FERMI ?

[kash]

walter can you tell what is KEPLER or FERMI ?

[Walter Roberson]

KEPLER

FERMI

[kash]

I have uploaded the image

http://imgur.com/JB7WiS1

[Walter Roberson]

And which parts of those are important to detect, and which parts of those are effectively "noise" for the purpose of your CBIR ?

For example, should the trees be treated as a key element, or should they be treated as transient because of the time scales involved, or should they be noted but they might need to be matched between different seasons with different appearances?

[kash]

Thing thing is that every element is to be treated and extract feature vector,no specific part,