How can I segment out the man made objects(such as buildings, water tank, etc) or remove trees and sky from the given images.

Answer 1

[Image Analyst]

You can use Photoshop, but even that will be tough. You can look here for published papers: http://www.visionbib.com/bibliography/contentscartog.html#Cartography,%20Aerial%20Images,%20Remote%20Sensing,%20Buildings,%20Roads,%20Terrain,%20ATR but I think most of those are for aerial images, not images like yours. There is nothing built in to MATLAB to do that. You may have to attempt your own algorithm, but I think you will not be able to do it - I know I couldn't for every possible snapshot that anyone could ever take. In fact, let me just say that it is impossible to do it automatically for all images one might encounter. And we don't do custom algorithm development here anyway.

[Mohan krishna Manchala]

Thanks for your reply. Can we use NVDI and egde based segmentation. I mean edges for trees are random but for building they are mostly regular lines or something like that. How to divide an Image into blocks(say 10*10) containing trees and not containing trees ? We can see this in the following images.

<<

<<

<<

>>

>>

>>

[Mohan krishna Manchala]

[Mohan krishna Manchala]

We can observe that edges of trees are random curve shape, but for buildings and other man made objects have horizontal or vertical or cross lines mostly. How to divide an Image into blocks(having tree or buliding)using egde shape in that block. I am trying to implement this paper. Kindly help me. http://ieeexplore.ieee.org/xpl/articleDetails.jsp?tp=&arnumber=4019932&url=http%253A%252F%252Fieeexplore.ieee.org%252Fxpls%252Fabs_all.jsp%253Farnumber%253D4019932

[Image Analyst]

That may be true for certain images, but as you can see it failed miserably for the first image you posted.

And if you have some large uniform area, how can you tell if it's a wall (man-made) or the sky or some other uniform natural scenery?

http://matlab.wikia.com/wiki/FAQ#Can_you_program_up_the_algorithm_in_this_article_for_me_and_explain_it_to_me.3F

[Mohan krishna Manchala]

True. I agree with you. Probably we can also use texture/color information of the large uniform area. However I need to remove trees and sky from image sequences similar to the ones posted above. So Can you help me out to do the following? If i have an edge detected image and have divided it into blocks of 10*10 pixels size, How to know whether a particular block has edges with 1. Varying slope, 2. horizontal edges, 3. vertical edges, 4. cross edges or 5.empty.

[Image Analyst]

What does "remove" mean to you? Images must remain rectangular. They can't have "ragged" edges. You can replace pixels with certain colors but you can't remove them and have the image remain an image that's recognizable to you as an image. You could put the pixels into a cell array but you can't display it like an image, and I don't think that's what you'd want anyway. See http://matlab.wikia.com/wiki/FAQ#What_is_a_cell_array.3F