How can I extract line features from a raster?

[jpdistef]

I'm new here so I hope I'm posting this in the right spot.

I'm trying to convert contour line features on a scanned topographic map and I'd love some help. My end goal here is to create a DEM for these maps. Here's what I've tried so far in ArcMap with corresponding images:

1. Georeference the map
2. Clip the frame and isolate one color band
3. I set each color value on the one band as a unique value on a new spectrum. This was in Unique Values under Symbology. I did this to identify the contour lines as foreground and the rest as background. I zoomed in so you can see the different colors of the black roads and blue rivers (green), the dark red/brown contour line (yellow) and the beige background (red)
4. I then grouped all unique values as black (foreground) and white (background)
5. I then used played around with the grouping and used ArcScan to clean this up a bit so I only got the bold contour lines and could remove the numbers. 
6. Finally, I converted it to a feature and got this!

You can see even in the super zoomed in section on file 6, it's rough, unconnected, and only includes the major contour lines. Any help would be INCREDIBLY appreciated. 🙂

[DavidPike]

I think it would be a nightmare imho.  Digitizing them would be a pain but there's going to be so much mandraulic work anyway to sort it out and then attribute the countours.   

[curtvprice]

What are you trying to accomplish?  For just about any use case in the U.S. I believe you would be better off (in my opinion) using elevation data available from 3DEP in The National Map.  The 3DEP is built from the best available data which in many cases are much more up to date and detailed than the contour data you are trying to capture. Even if the only data are available are contours, USGS creates their DEMs hydro-enforced, which is a great deal of work beyond just capturing the contour lines.

[jpdistef]

The map attached was produced in the 1890s.

This may be naïve, but I'm trying to create DEMs for multiple time periods in one watershed using the historical topo maps produced by the USGS, then do a DoD analysis to identify at major geomorphic changes in the watershed throughout its history. I imagine the changes will be too small to notice outside of margins of errors, but I figured it can try!

Is DavidPike right; is this just going to be a nightmare?

[curtvprice]

I found a paper about this. It sounds like it doable but problematic, remember that you are probably talking about three different methods of creating contours (with different types and amount of error on each). And -- very small changes with respect to the typical USGS quad contour interval.

https://www.sciencedirect.com/science/article/abs/pii/S0169555X11002923 

Yes, it's a lot of work to extract contours in an automated way, and very image-specific. Usually the least painful way to do it is to digitize it 'by hand' - though I have to say this is much easier using modern tools than on the old big digitizing tablets!

[curtvprice]

Have you compared the contours of your geo-registered quads? If there was anything major it would probably be visible doing a visual comparison of contours.
Honestly I wouldn't expect the landscape to change that much over 100 years to exceed the error between the different methods of creating contours.

[jpdistef]

Looking at the paper you attached, it looks like total vertical uncertainty ranged from about 1-10 meters. Looking at one interesting section map and comparing it to the geo-registered quads, as well as another topo map of the region from the 1950s, there appears to be some noticeable change, but it's rather small. Sometimes within the uncertainty range of 1-10 meters, sometimes not.

[JayantaPoddar]

I think you need to do a Heads-up (Manual) digitization and attribution here, unless generating the contour from a DEM is an option.

To generate contours from a DEM, use Contour  > Choose appropriate Contour_Interval and Base_Contour.