bUnwarpJ RANSAC question

[Anderson, Charles (DNR)]

I have many images of fish on a measuring board, where the camera position is only approximately above the fish and extraneous landscape is in view outside the board.  I'd like to register the fish source images against a target image of the measuring board using only landmarks on the measuring board. An outside mask could exclude the landscape and an inside mask could exclude the fish. I have planned to use a blue board with red dots as landmarks in order to use color-split to create the masks (and potentially centers of the landmarks).

Would the landmark extraction and RANSAC matching be able to handle the red dots on blue, or would the lack of gradient and orientation information preclude use?

Or is there an alternative RANSAC code available that could match using only (x, y) coordinates of landmark points?

Or is there a better way to build landmarks into the board that work with SIFT or bUnwarpJ description and still allow color separation to create inner and outer masks?

Thanks for suggestions.

Charles Anderson
Minnesota DNR Fisheries

[Stephan Saalfeld]

Hi Charles,

if the measuring board has sufficiently distinctive texture, you will
not need to mask anything.  When you do feature extraction, the RANSAC
filter will jump on the largest set of corresponding features supporting
a transformation of the given class (choose perspective if your camera
tilts slightly).

Can you share some images?  Then I can give more educated advice.

Best,
Stephan




On Tue, 2011-02-22 at 13:37 -0600, Anderson, Charles (DNR) wrote:

> I have many images of fish on a measuring board, where the camera position is only approximately above the fish and extraneous landscape is in view outside the board.  I'd like to register the fish source images against a target image of the measuring board using only landmarks on the measuring board. An outside mask could exclude the landscape and an inside mask could exclude the fish. I have planned to use a blue board with red dots as landmarks in order to use color-split to create the masks (and potentially centers of the landmarks).
>
> Would the landmark extraction and RANSAC matching be able to handle the red dots on blue, or would the lack of gradient and orientation information preclude use?
>
> Or is there an alternative RANSAC code available that could match using only (x, y) coordinates of landmark points?
>
> Or is there a better way to build landmarks into the board that work with SIFT or bUnwarpJ description and still allow color separation to create inner and outer masks?
>
> Thanks for suggestions.
>
> Charles Anderson
> Minnesota DNR Fisheries