Seeking clarification re: Georeferencing/Orthorectification workflows

Question

Hi all,

I am hoping some folks with more experience and expertise can help clarify the best path forward for me:

I have multiple sets of scanned historic aerial images (1920s - 1950s) that I need to georeference. For the set I am working with now, there are ~200 scanned images I want to make into a seamless mosaic image after georeferencing. The scanned images include the whitespace from the pages around the edges of the images.

My workflow so far has been to load a tile, georeference with ~6 control points, Clip Raster to remove the whitespace border, and repeat. I have noticed that the edge alignment of resulting images is not always perfect - there are disjointed/offset edges where distinctive features don't align. The sidelap and overlap of the tiles is not fantastic - maybe 12-25%, varying by side and image - so I understand that the best possible outcome may not be perfect. I recently tried out mosaicking a few tiles together hoping that the "Blend" operation would help smooth the edges, but it did not.

I would like to ensure that I can produce the most seamless mosaic possible, so I have started to wonder if my workflow should be different - should I (or can I) be trying to orthorectify these images rather than just georeferencing? Should I clip the borders, mosaic/reference all the images to each other first to get more seamless edges, and then georeference the larger mosaic? What process and tools would you use to make the best possible seamless, georeferenced mosaic image down the line? Am I completely overlooking any steps/tools?

Any suggestions or insights are greatly appreciated! Thanks in advance!

MaxPetri · Answer

Sorry it took me this long to get back, I need to check the replies more often.

I have unsuccessfully tried to apply the footprint before georeffing; by JOINing ungeoref'd photos to a mosaic dataset that has the general shape of a footprint from an other source, but I don't think that works because the mosaic dataset is reliant on georef'd photos. Sorry, no help there. . Ideally you georef all your photos and then load them all into a mosaic dataset.

Cropping should work, but folks may over crop, which could reduce the amount of overlap needed to get a good SEAMLINE to run between your FOOTPRINTs. You'd also be losing the fiducials present at the edges and corners of your photos if you cropped, the fiducials are required for orthorectification and ortho-mosaicing.

You could also play with the different transformation settings to get a "better" georeference, ie, 2nd poly or spline. Each of these will stretch and squish each photo to a varying degree, sometimes too much depending on where GCPs are placed.

JasperRomero · Answer

Thanks for weighing in @MaxPetri - sounds like we are definitely working on something similar.

I am not too familiar with the footprints tool/settings, or seamlines. Based on your description, I will get familiar with both of these. I have been manually cropping the borders in Pro, but if there's a way to avoid that step I am interested! However, I typically crop before georeferencing to make it easier to align the seams between pairs of images - can the footprints be applied before the georeferencing?

MaxPetri · Answer

I've been working on a process to do something similar, where I am not orthorectifying the mosaic, but just creating a mosaic of the georeferenced photos. When working with historical aerial imagery you may or may not have the collars cropped off of the images. This process might work with either. You will want to know the general characteristics of the color depth of the images to account for more or less black or white pixels, especially the collars, since they would have full black or full white edges. Hopefully the build footprints tool can mostly ID those, and use the footprints to essentially mask the collars of each photo, then use the seamlines to do the seaming between photos. Ideally, if you are seeing misalignments, you'd re-georef that photo to align with it's neighbor.

1. Create a new mosaic dataset

2. load the georeferenced images in

3. build the footprints using: radiometry | minimum data value: 1(or the lightest black color in the collar) | maximum data value: 254 (or the darkest white in the collar) | approximate number of vertices: 10 (to reduce the amount of aberrant shapes, we need it to be close to rectangle as possible) | minimum region size: 1000 (to account for clouds or other features that might confuse the function, this number may be dependent on the spatial resolution of your photos).

4. once your footprints are done, you can EDIT any individual records in the footprint to make sure they align with the image.

5. build seamlines on default settings.

6. export or view your mosaic dataset.

With this method, you may not need to pre-crop your images, just georeference and make sure adjacent photos visually align.

GRAN-AJU-1N-182.jpg

GRAN-AJU-1N-182(cropped).jpg