I have a grid of N x M images of equal size which I want to "stitch" into one large image. Within the X or Y stitching dimension each image overlaps with the next one by a constant but unknown amount of pixels. If this overlap range would be known the position of each image in the final one could simply be calculated.
I wonder if the overlap can be identified using a combination of standard IGOR image analysis or matrix math tools, but currently I don't see where to start. Any suggestions are highly appreciated!
You can compute the overlap of any pair of images to sub-pixel resolution.
Lets assume that you have two waves A and B that share a few pixels on the right hand side of A and the left hand side of B. Start by estimating the overlap from manual inspection. Suppose that overlap is roughly W. The next step is for you to duplicate W pixels from the right hand side of A and W pixels from the left hand side of B and use ImageRegistration. The output parameters of the operation will give you the precise overlap and any stretching factor.
Please note that this approach is, in general, not appropriate for stitching arbitrary panoramas because it does not handle blending and is designed for handling image pairs only.
thanks! I was looking for something like this. It seems that the initial estimate of the overlap needs to be fairly close to the actual one. That should work in my case since the overlap is always roughly 10% of the X or Y dimension.
You can compute the overlap of any pair of images to sub-pixel resolution.
Lets assume that you have two waves A and B that share a few pixels on the right hand side of A and the left hand side of B. Start by estimating the overlap from manual inspection. Suppose that overlap is roughly W. The next step is for you to duplicate W pixels from the right hand side of A and W pixels from the left hand side of B and use ImageRegistration. The output parameters of the operation will give you the precise overlap and any stretching factor.
Please note that this approach is, in general, not appropriate for stitching arbitrary panoramas because it does not handle blending and is designed for handling image pairs only.
A.G.
WaveMetrics, Inc.
April 8, 2013 at 10:01 am - Permalink
Hi A.G.,
thanks! I was looking for something like this. It seems that the initial estimate of the overlap needs to be fairly close to the actual one. That should work in my case since the overlap is always roughly 10% of the X or Y dimension.
Cheers
C
April 9, 2013 at 01:12 am - Permalink