Subscribe to this thread
Home - General / All posts - Manifold 8, help needed to register 1:50000 Gabon map raster
Corentin50 post(s)
#08-Oct-19 20:33

Hi,

I bought a set of 1:50000 Gabon map issued by french IGN (Institut National de Geographie), they are image in JP2 format, they do give information about ellipsoid, projection, graticule and grid, but I am unable to precisely registering them, I always have about 200m shift if I compare with google earth image. This range of precision matters since I need these maps to locate hydroelectric potential sites.

The legend says : Universal Transverse Mercator zone 32N, ellipsoid of Clark 1880

Here is what I do :

1. create a new drawing, assign projection of these drawing in UTM 32N, Clark 1880 IGN (cf image attached)

2. put on this drawing as many points as the graticule let me do so (e.g 16 point with 5min space)

3. export the image in png, import it back to decompress the data (in order to get more registration methods)

4. put the corresponding control points both on the drawing and on the image

5. register (I tried almost everything : affine (scale shift, rotate), with and without scale pixel equally in X and Y direction, triangulation, and affine (scale shift))

When I control the result, the status bar gives me both correct lat/long and projected coordinate (cf file status bar.jpg)

When I create a map with both my control drawing and my image things seems ok, I have almost no shift (shift is below 10m which is smaller than the size of my pixels, cf image attached)

When I control with google earth, things goes wrong, the last image shows in purple a river roughtly traced on my control drawing from google earth, there is a shift of 250m :(

What is wrong?

The set of map is quite old (1960-1980) but I have a colleague who succeed doing the same thing with maps of Benin from IGN... Is it Google earth that is not accurate, are the IGN map not accurate? Something went wrong in my registering process?

There are other ellipsoid Clark 1880 in the projection database, I tried one but it gives the same result

I have 50 of these maps, it costed me about 350EUR, it is very deceitfull that I cannot register them with manifold...

PS I haven't tried the numeric method for registration, takes too long with my old computer (>1hr)

Attachments:
250m Shift.JPG
map control drawing - registered image.JPG
status bar.JPG
UTM 32N Clark 1880.JPG

oeaulong

218 post(s)
#08-Oct-19 22:13

I think it will be difficult to help without an example of one of the maps. I have put in a request to download two maps in this area. I do not expect to hear a response for another 9 hours or more.

As I am trying to work through your workflow and understand your issue. I ran a test and generated a grid of graticule points in both UTM32N;WGS84 (auto) and in UTM32N;Clarke1880. Then in a map overlaying these I looked at the offsets of the points. I see you are using a reference origin of Lat:0.0, Long:9.0 degrees. My past experience tells me that a 0,0 origin might suit better as a starting reference for adjusting projection. Changes in these values can either increase the precision if it falls within or close to your mapping area, but it can also generate wild changes. I suspect this may be a source of your frustration. I haven't been able to reproduce 250meter offsets so cannot in real detail address your issue.

Regarding Google Earth and accuracy. The Mercator; WGS84 (auto) projection in use for Google earth in version 8 is not necessarily good for maintaining highly accurate precision. In Mfd9, this is now the very common web mapping standard, WGS / Pseudo-Mercator EPSG:3857. For Presentation, it usually suffices as the GoogleEarth image server and Bing/Microsoft image server provides this. In the US where I reside, using UTM is common, but still there is scanned map data in NAD 27, NAD 83, and others. I have had some success with using UTM in a precise fashion, but it does take some care.

So the question is, to me, What is it you are trying to accomplish in the bigger picture.?

If, you are trying to accomplish mapping the older image to a newer datum, such as WGS84 then I think you would need to import the JP2 IGN Gabon image, then register graticule/grid points from the image to matching ones in a drawing/map component with generated graticule/grid points in the newer projection & datum combination appropriate for your analysis area. This may be another projection/datum combination aside from the two referenced here. I would read up on the creating a graticule as objects subject here:

This will give you exact points in a drawing of the target projection/datum that can be used to georeference the image to.

Though you are using UTM as a reference, the datum will make much differences in how things get place on the spheroid defined. Remaining in Clarke 1880, would in my opinion be unwise, though may be necessary to create the original library image that ultimately gets reprojected with 50 images to complete.

Perhaps someone else will have more concrete advice for you. I will respond if I can reproduce your offset and how I might attack a fix.

drtees77 post(s)
#08-Oct-19 23:45

Not certain if any of this will help, but here is how I typically deal with unprojected raster images. First, do not assign a projection to the image. If the georegistration process works as planned, the image will be assigned the same projection as the reference drawing. Also, exporting the image to png may not be necessary. With the image open, use Image/convert to and select RGB. This will uncompress the JP2 format and allow you to directly modify the image's projection.

The next step is to choose a drawing or projected image that has recognizable locations that also occur on the image you want to register. I am assuming that you have latitude and longitude lines on the image to be registered. I look for sharp corners or intersecting lines that occur on both the target and the source.

I find that the default registration (affine) works well most of the time. Triangulation or numerical requires extremely careful placement of control points. Triangulation will delete any part of the image that is outside of the control points, but every control point on the source will fall exactly on the target. I have never had much luck with numeric. I typically end up with a completely distorted image, probably because I am only interested in registering a fairly small portion of a larger image.

geozap
131 post(s)
#09-Oct-19 03:39

Probably your issue is caused by not appropriate datum shift parameters. Clark (or any other ellipsoid) to WGS84 shift parameters are different for each part of the world. So you do the registration correctly, but when combining the registered data with a drawing/image in another projection (such as google earth maps) they don't match because the Gabon maps are not transformed to the other projection the best possible way.

So maybe you should define a custom datum with the best shift values for Gabon. I attach some pdfs that may help you to find those parameters.

Attachments:
draft_appa.pdf
GPS.G1-X-00006.pdf
S60_Ed3Eng.pdf
wgs84fin.pdf

geozap
131 post(s)
#09-Oct-19 17:10

Adding to my comment, could that be the maps' datum: https://epsg.io/4266 ?

"Plain" Clark ellipsoid (plain=without shift to WGS84), would be strange to be the maps' datum.

tjhb

8,883 post(s)
online
#09-Oct-19 08:48

I haven't tested yet, but like others I will give my unfounded 2c worth. Briefer.

1. I think you are doing everything right. It looks good.

2. Don't underestimate Google's capacity to fudge geography, where they have little or no capacity to earn income from advertising.

3. I suspect Google may have bought or borrowed IGF data, but when they used it they forgot about non-US datums. (That would be the 200-odd metre shift.)

4. The main point is to triangulate. A third data source is essential. Even for Gabon, there are many. Mostly global, it must be said, but that does not make them less worthwhile.

5. Google is not any kind of standard. If you want a standard geographic reference, use data from FAO, USGS, NASA or ESA.

6. Point 5 is really important. Google is not any kind of standard. Only a selective convenience.

Dimitri


5,552 post(s)
#09-Oct-19 09:55

I bought a set of 1:50000 Gabon map issued by french IGN (Institut National de Geographie), they are image in JP2 format, they do give information about ellipsoid, projection, graticule and grid,

No need to georegister if IGN provides correct info. Here is what I recommend:

1. Import the image.

2. Assign the projection using the parameters IGN says you should use.

If you do the above, what happens? If you feel the result of the above is wrong, then:

1. Say exactly what is wrong and

2. Post a link to the original jpeg file you got from IGN and the original information IGN says should be used for that file.

Corentin50 post(s)
#09-Oct-19 12:32

Thank you for your answers,

In fact some of the image are already georegistered, and yet they have this 200m shift compared to google earth. see attached snapshots. If I try to register myself the image it does not resolve the pb.

Here is a link to download an extract of the data

https://www.dropbox.com/s/e6s4jb6ojfr37pi/Trying%20registering%20Gabon%20map.map?dl=0

Attachments:
River + google earth.JPG
River+ IGN.JPG

Dimitri


5,552 post(s)
#09-Oct-19 13:47

Here is a link to download an extract of the data

Unfortunately, that is not the original jpeg file.

The way to investigate such situations is to begin with the original, unmodified file from IGN, including any accessory files provided by IGN, together with all the original information that IGN provides about the projection / coordinate system used by that file.

By the way... don't use + characters in file names you attach. It's best to stick to alphanumeric characters and underscores and to avoid special characters.

Corentin50 post(s)
#09-Oct-19 14:23

The file I have attached has been obtained simply by opening manifold doing image/import select the JP2 file from IGN. That is all I have done, I added the control point,but I have not registered the image myself.

I would really like to share with you the original file, but I cannot. that is why I deleted most of the pixel in the image in manifold, but I let everything that could help : lat/long line, legend, information about the projection, north direction, the river I am intersted in getting a precise position...

Attachments:
River google earth.JPG
River IGN.JPG

Dimitri


5,552 post(s)
#09-Oct-19 20:14

The two screenshots indicate that the image you imported from the jpeg does not agree completely with what the Google satellite view shows. The two are not wildly off. It's not like one of them is appearing as a microscopic dot off the coast of Africa in the ocean.

OK. That means one of three possibilities:

1. Google is off.

2. The IGN image you imported is off.

3. Both Google and the IGN image are off.

Without the original IGN image, together with all accessory files, plus all the original information that IGN provides we can't say whether the IGN image is off, or if some manual adjustment needs to be done to the import (tinkering with the datum, etc.) to get it exactly on.

With the info requested, we could import it into 9 and tell you in a couple of minutes whether it was on/off or if off a possible route to get it spot on. I wouldn't just compare it to Google, I'd compare it to many other data sources as well.

But... as other comments in this thread have indicated, that the IGN image does not agree with Google is not proof there is something wrong with the IGN image.

If you want to see that for yourself, download Viewer (it's free) and take a look at the same region using Google, Bing, OSM and other layers for backgrounds. None of the web servers agree exactly with each other. Import the IGN image as well and see how it lines up in Viewer.

Manifold User Community Use Agreement Copyright (C) 2007-2017 Manifold Software Limited. All rights reserved.