Historical Image Processing - Farm Field Irregularities - Elevation/Altitude issues


I’m processing historical imagery and am noticing a lot of elevation artifacts primarily in farm areas where the crops create similar geometric patterns. 

I’ve tried a number of different processing settings, but haven’t made an impact on having fewer artifacts.

I’m interested in strategies I could try in automated processing, or am I reduced to correcting the points by hand?

Thanks :slight_smile:

Hi Steven,

Have you tried to process with the Alternativecalibration method? It can be used only if you have images’ coordinates and usually, gives better results for flat, uniform areas such as an agricultural field.

Also, adding MTPs, normally, helps.

Feel free to post some screenshots from the quality report of your project, so that I can have a better understanding of the case and be more useful :slight_smile:

Are you working from analog imagery?

If so, what is the digital transfer res and can you increase the pixel density by rescanning to fine pixel pitch?

Hi Christina,

Thanks for your reply, I haven’t tried the Alternative calibration method yet. Sections of the area have elevation change, and my understanding is that that method may not handle that well. Additionally it can be a challenge actually finding the image center in present day to give it an accurate coordinate. I’ll give it a shot though. :slight_smile:

For adding MTPs would that be for each affected area, or just add one in a cluster of affected areas?

I’ll try out your suggestions :slight_smile:


Hi Gary,

Yes analog imagery, the imagery was scanned at 1200 dpi and provided to me. Each pixel represents about 50-75cm on the ground.

I’m not sure if I would be able to get it rescanned, but if I was is there a dpi that you would recommend for historical imagery such as this?

Cheers :slight_smile:

To improve the relative accuracy of a project by adding MTPs, you need to add MTPs on each affected area: 

You should click some points of the rayCloud and then, check their reprojection on the input images. The reprojection is good when the point that you click in the rayCloud is displayed at the same location on the images:

If the reprojection is good, then the relative accuracy of the model is good. If not, you should add some MTPs and Reoptimize: Add MTPs.

Thanks Christina, that makes sense. I’ll investigate further as to which parts of the image the tie points are matching to in bad areas. If they are matched incorrectly is it better to add new MTPs in that area or to adjust/reclassify the erroneous errors… or a combination of both.

As you can see from below, I have wide swaths that appear to work out okay. But the area is also filled with farms and they seem to be creating all sorts of problems :slight_smile:


Steve , I think you and I attended the Pix4D class in ATL together. If thats the case , hi again!

Christina’s suggestion is the most rational approach, I worry that the raw data may not have enough pixel contrast for Pix4D to extract from, so some other image enhancement process prior to Pix4D might help.

Just 1.25 cents worth anyway…hehehe…sorry!

Let us know how this is going. I believe that MTPs will help. Let´s see… :slight_smile:
I would also give it a try with Alternative.

Feel free to post some screenshots from the quality report, so that we can check whether the processing options are OK.

Hi Gary, nice seeing you again :slight_smile:

I am considering trying to get them rescanned to a higher DPI to try and provide more contrast. I’m unsure of what sort of additional image enhancements to try at this point, but will potentially reevaluate.

Thanks :slight_smile:

Hi Christina,

Thanks, I haven’t had a chance to try but will evaluate MTPs soon.


Hi Christina,

I still haven’t gotten to looking at the actual processing, but just wanted to confirm that adding more GCPs wouldn’t really do anything in this situation correct? 

Would it even be worth having them in the project at all while these DSM issues are occurring? To me they almost seem like the last thing to get sorted out…

Really to me I should just focus the effort on tests with some additional manual tie points, a session with the setting you reference, potentially rescanning the imagery at a higher resolution or doing some form of source image manipulation like histogram stretching or sharpening to improve the contrast etc. Would you agree?



Hi Steven,

Adding more GCPs is not a bad idea. It could certainly help if the GCPs are added in the “problematic” areas.

However, before going to the field to get the GCPs, I would recommend you to process the dataset with Alternative and several MTPs.

Please note that the Alternative option can only be used when the images are geolocated (they have coordinates). If this is not the case, unfortunately, you will not be able to try this option out.


1 Like

Hi Christina,

The manual tie points seemed to work, I tried the Alternative setting and it produced different results I can’t really ascertain whether they’re better or worse as they also have a lot of artifacts. What accuracy should the image center coordinates be to be used in a solution.

A separate issue I’m having is that I can’t get my horizontal image accuracy down to a reasonable level. I’d like below 10m but in some areas of the project it’s closer to 40m or more.

Would it be possible to arrange a screenshare to go through some of my challenges? I feel like there’s something I’m missing.