[GRASS5] raster georectification

GRASS GRASS developer
1

I have a set of scanned maps that I want to georectify. However I am unsure of the most efficient way to handle the three images (three color bands) associated with each image. I have imported them into a temporary xy location and can display with d.rgb (takes about 10 seconds to display). I have grouped all with i.group, and used i.target to set up new target location and mapset. Now I am ready to use i.points. My question is this: Since it looks like all three element maps (rgb) need to have i.points run with associated coordinates, should the three elements of the color table me merged first in order to reduce the number of maps that need georeferencing, or is assigning coordinates to all three bands the standard approach? I read in "Open Source GIS: A Grass GIS Approach (Neteler and Mitsova) that one of the reasons that scanned color maps are broken into color bands is to reduce computational overhead and there by speed up image processing. Does it make sence to merge these maps first in order to reduce number of maps that need georectifying, or would I be more than paying for that time savings with very slow image processing?

btw... using grass5.7

Thanks

Kirk
------------------------------------------------------------------------
Kirk R. Wythers tel: 612.625.2261
Dept. of Forest Resources fax: 612.625.5212
University of Minnesota email: kwythers@umn.edu
------------------------------------------------------------------------

2

Kirk R. Wythers said:

I have a set of scanned maps that I want to georectify. However I am
unsure of the most efficient way to handle the three images (three
color bands) associated with each image. I have imported them into a
temporary xy location and can display with d.rgb (takes about 10
seconds to display). I have grouped all with i.group, and used i.target
to set up new target location and mapset. Now I am ready to use
i.points. My question is this: Since it looks like all three element
maps (rgb) need to have i.points run with associated coordinates,

Why do you need to run i.points three times ? Since they are grouped and
i.points and i.rectify work by group (you can determine which images in a
group you wish to rectify), you only have to use i.points once for
rectifying all images in a group.

should the three elements of the color table me merged first in order
to reduce the number of maps that need georeferencing, or is assigning
coordinates to all three bands the standard approach? I read in "Open
Source GIS: A Grass GIS Approach (Neteler and Mitsova) that one of the
reasons that scanned color maps are broken into color bands is to
reduce computational overhead and there by speed up image processing.
Does it make sence to merge these maps first in order to reduce number
of maps that need georectifying, or would I be more than paying for
that time savings with very slow image processing?

It depends of what kind of processing you wish to do and what type of
image you have. Often one uses a specific band for specific analysis or
you can calculate indicators using the different bands, so it is generally
not advisable to merge them, not for speed reasons but for analytic
reasons.

However the merged image could help you with i.points since sometimes it
is easier to identify a point on a merged image than on an individual
band.

Moritz

3

On Jan 7, 2004, at 10:57 AM, Moritz Lennert wrote:

Why do you need to run i.points three times ? Since they are grouped and
i.points and i.rectify work by group (you can determine which images in a
group you wish to rectify), you only have to use i.points once for
rectifying all images in a group.

Perhaps I am misusing i.group. I started with 6 scanned maps, each of a different geographic space representing adjacent townships and ranges (essentially a 3X2 matrix of a single larger map). After importing into GRASS, I was left with 3 color bands for each map (rgb). All 6 maps need to be assembled into a single map with all color bands. I defined a single group with i.group. I now believe that I should have defined 6 separate groups with i.group (one for each of the 6 geographic locations). So the question is: in what order should these steps be completed?

1. merge rgb bands into a single map for each of the 3 bands
2. georectify with i.points
3. overlay adjacent maps

It depends of what kind of processing you wish to do and what type of
image you have.

These maps are hand drawn forest with color pencils (nothing fancy but very detailed and of historic significance). I don't want to loose the "wow" effect of looking at all the colors together until they get heads up vectorized.

------------------------------------------------------------------------
Kirk R. Wythers tel: 612.625.2261
Dept. of Forest Resources fax: 612.625.5212
University of Minnesota email: kwythers@umn.edu
------------------------------------------------------------------------

4

Kirk R. Wythers wrote:

> Why do you need to run i.points three times ? Since they are grouped
> and
> i.points and i.rectify work by group (you can determine which images
> in a
> group you wish to rectify), you only have to use i.points once for
> rectifying all images in a group.

Perhaps I am misusing i.group. I started with 6 scanned maps, each of a
different geographic space representing adjacent townships and ranges
(essentially a 3X2 matrix of a single larger map). After importing into
GRASS, I was left with 3 color bands for each map (rgb). All 6 maps
need to be assembled into a single map with all color bands. I defined
a single group with i.group. I now believe that I should have defined 6
separate groups with i.group (one for each of the 6 geographic
locations). So the question is: in what order should these steps be
completed?

1. merge rgb bands into a single map for each of the 3 bands
2. georectify with i.points
3. overlay adjacent maps

If the 6 individual maps are just "tiles" of a single map, the logical
sequence is to patch them together first, then rectify the single
large map. Otherwise (i.e. if the individual maps use different
projections), you need to rectify them separately then patch the
results together.

The point which Moritz was making was that you don't need to rectify
the individual R/G/B bands separately. Each run of i.rectify will
simultaneously rectify all three bands (assuming that they have been
grouped).

You shouldn't actually need to merge the R/G/B bands at any point, and
doing so isn't desirable in most cases. When generating a composite
map, you usually have to reduce the colour depth, otherwise you end up
with massive colour tables which tend to make most operations
extremely slow (either that, or they just fail due to insufficient
memory).

The main situation where generating a composite map makes sense is for
a "preview" (particularly for programs which only allow the use of a
single image rather than separate channels).

E.g. you might generate a composite map (with a low colour depth) to
use when marking points in i.points, but you would actually rectify
the separate R/G/B channels (you might also wish to rectify the
composite map, or you could generate a new composite map from the
rectified channels).

--
Glynn Clements <glynn.clements@virgin.net>