algorithm - How to separate object from background in an image? -

I have a block image, looks like this:

// Posting pictures to me Is not allowed because I am a new member, so only the link:

// I can not post two hyperlinks, so I am only posting the link to the map file.

> Edit: I believe I can now post images:

I have a map There is also a file which clearly shows segments:

Now, what I need to do is to create a binary image file that only deer white paint in the center, the rest are painted Is black colored

What methods do you suggest for merging?

There is something like this in my mind:

1. Calculate the color average for each segment.
2. Compare them and merge the most common sections.

If I do this, then I end up with 3 sections: floors (white part), wall (black and light brown part joint) and object (gray part).

What could be done at this point to get the object properly?

Note that the object does not have to be in the center, it can also be partially off-screen.

(I also thought about counting the area takes each segment and labeled the smallest area in the form of the object; but sometimes it may be that when most objects are covered in the objects , So it can not show the correct result.)

I really appreciate any help. thank you in advanced.

This is a difficult question because "object" is a subjective term. Clearly, you want the most interesting thing, so we have to decide what an interesting object looks like. It must be some statistically.

Let's say that, like your images, the object of your interest is one of a small number. We will only calculate one digit for each segment, and call the object the highest scoring.

I'll just play together adding different score functions simultaneously. Some good people can be:

• Distance of center or square distance of a segment (this is your example object) for the exact center of the image.
• The pixels of The numbers are at the limit of the image of the segment (this will find your example object because the poor cluster has large limitations)
• The number of side borders divided by inside in pixels The number of, if you think your object "More interesting" than in the background
• The number of SIFT keypoints divided by the number of pixels, as the previous
• Brake rb space is in a relatively small number of compartments, such as 512 ( If you are working with 8-bit images, then break each of R, G, B in 0-31, 32-63 etc.). Look at the properties of each section considered as distribution in this place.

An interesting object may have high-entropy distribution, or the low-entropy one can be.